Code Review Report

Project: BofAI Documentation (Docusaurus)
PR: main -> ai-bankofai-patch-1
Review Date: 2026-04-06
Reviewer: AI Code Reviewer (Code Review Skill v1.0.0)

PR Overview

Branch Information

Commit History (Recent)

Note: The commit history shows significant churn: multiple rename-undo-rename cycles for model files (chatgpt-5 → gpt-5 → chatgpt-5 → gpt-5) suggest an editorial process that could benefit from cleaner, squashed commits before merge.

Review Summary

Verdict

Verdict: Request Changes

Findings at a Glance

Summary

This PR is a documentation-only update covering four logical areas: a substantial API reference rewrite, new Memory Service documentation, model renaming (ChatGPT-5.x → GPT-5.x) with three new GPT-5.4 model pages, and pricing/sidebar updates. The overall quality is good — the restructured API reference is significantly more developer-friendly with inline request/response examples, and the new model pages are thorough.

However, three major correctness issues require fixes before merge: a pricing calculation example in the English docs uses stale per-token rates that produce a wrong credit total; a pricing number within the same GPT-5.4 Nano file is internally self-contradictory; and the API base URL domain change (.com → .io) appears with no migration notice, which would be a breaking change for existing API consumers.

Five minor issues also need attention: missing top-level H1 page titles in six model/pricing files (breaking Docusaurus page rendering), trailing whitespace in sidebar config files, a broken/incomplete pricing link sentence, and inconsistent pricing unit notation across files.

Change Summary

Group 1 — API Reference Overhaul

Purpose: Improve developer experience by replacing terse schema-table references with concrete request/response JSON examples, and expand the Messages endpoint docs with full Claude-compatible field documentation.

Group 2 — Memory Service Documentation (New)

Purpose: Document a new Memory persistence feature that gives agents long-term context across sessions.

Group 3 — Model Renaming & New GPT-5.4 Pages

Purpose: Rebrand GPT-5.x models from "ChatGPT-5.x" to "GPT-5.x" naming and document three new GPT-5.4 variants released in March 2026.

Group 4 — Pricing & Usage Update

Purpose: Reflect new model offerings (GPT-5.4 variants, MiniMax-M2.5, Kimi-K2.5, GLM-5), update deposit language to TRON/TronLink specifics, and improve readability.

Detailed Findings

Major

[MJ-01] Pricing Calculation Example Uses Wrong Input/Output Rates (EN)

Description

The English pricing calculation example references GPT-5.2 but uses per-token rates that no longer exist in the pricing table. The table correctly lists GPT-5.2 at 1.75 input / 14.00 output, but the example formula uses 1.25 and 10 — leftover values from the old generic example — producing an incorrect result of 512.5 credits. The correct calculation should yield 717.5 credits.

Notably, the Chinese (i18n) version of this same file was updated correctly to 717.5 credits using 1.75/14 rates, making this an English-only regression that creates a factual inconsistency between the two language editions.

Code

For example, if you use GPT-5.2 to ask a question (10 input tokens) and the AI responds with
an answer (50 output tokens), the entire dialogue consumes 512.5 credits
(calculated as: 10 × 1.25 + 50 × 10).

Recommendation

For example, if you use GPT-5.2 to ask a question (10 input tokens) and the AI responds with
an answer (50 output tokens), the entire dialogue consumes 717.5 credits
(calculated as: 10 × 1.75 + 50 × 14.00).

[MJ-02] GPT-5.4 Nano Output Pricing Self-Contradictory Within Same File

Description

The GPT-5.4 Nano model page states the output price as $1.25/1M tokens in two places (the overview paragraph and the Key Features bullet), but the Credits and Pricing table at the bottom of the same file shows $1.20 output. The two values disagree within the same file, and users relying on either will have inaccurate cost expectations.

The pricing table in pricing-and-usage.md also shows $1.20, suggesting $1.20 is the intended value and the prose is the error.

Code

<!-- Overview paragraph (line ~4) -->
At just $0.20/$1.25 per MTok, it delivers reasoning capabilities…

<!-- Key Features bullet (line ~8) -->
- **Extreme Cost Efficiency**: $0.20/1M input + $1.25/1M output …

<!-- Credits and Pricing table (line ~34) -->
| GPT-5.4 Nano | $0.20 | $1.2 |

Recommendation

Unify to a single value (verify against the authoritative source). If $1.20 is correct, update Overview and Key Features prose:

At just $0.20/$1.20 per MTok, …
- **Extreme Cost Efficiency**: $0.20/1M input + $1.20/1M output …
| GPT-5.4 Nano | $0.20 | $1.20 |

[MJ-03] API Base URL Domain Change With No Migration Notice

Description

The API Base URL was silently changed from https://api.bankofai.com to https://api.bankofai.io. This is a breaking change for any existing API consumers who have the old URL hardcoded. The diff removes the old URL entirely with no deprecation notice, redirect information, or migration timeline. Developers who are actively using the .com endpoint will receive no warning.

Code

-| `https://api.bankofai.com` | Production |
+- **Base URL:** `https://api.bankofai.io`

Recommendation

Add a migration notice (inline callout or a dedicated section) informing users of the URL change:

> **Migration Notice:** The production API base URL has changed from
> `https://api.bankofai.com` to `https://api.bankofai.io`.
> The old domain will continue to be supported until [date]. Please update
> your integrations accordingly.

Minor

[MN-01] H1 Page Titles Removed From Six Documentation Files

Description

The three renamed model files had their H1 (# Title) headers removed (replaced with a blank line), the three new model files were never given H1 headers, and pricing-and-usage.md had its # Pricing and Usage H1 removed. Docusaurus uses the H1 heading to set the browser page <title>, the breadcrumb label, and certain navigation fallback labels. Without H1, pages may render with blank titles or fall back to the sidebar label — behavior that varies by Docusaurus version and theme.

Code (example — gpt-5-mini.md)

-# ChatGPT-5-mini
+

 ## Overview

Recommendation

Restore or add H1 headings that reflect the new model names:

# GPT-5-mini

## Overview
…

For pricing-and-usage.md:

# Credits & Pricing

[MN-02] Incomplete/Broken Link Placeholder in Pricing Page (EN)

Description

The sentence "For specific model pricing, please check here：" ends with a Chinese full-width colon ： followed immediately by the pricing table. There is no hyperlink target. This looks like an unfinished edit where a link was intended but never added. The Chinese full-width colon also looks out of place in the English version of the file.

Code

…Some models do not support web search marked as "-". For specific model pricing, please check here：

| Model             | Input (Credits/Token) | …

Recommendation

Either add the intended link or remove the trailing phrase:

…Some models do not support web search marked as "-".

| Model             | Input (Credits/Token) | …

[MN-03] Inconsistent Pricing Unit Notation Across Files

Description

Two different pricing unit systems are used across the changed files without cross-referencing:

• pricing-and-usage.md uses Credits/Token (e.g., GPT-5.2: 1.75 / 14.00)
• New model pages (gpt-5-4*.md) use $ per 1M tokens (e.g., GPT-5.4: $2.50 / $15.00)

While the numeric values appear to be consistent (the Credit values align with dollar values), the notation is entirely different. A user reading a model page and then the pricing page will encounter two measurement systems with no explanation of equivalence.

Recommendation

Adopt a consistent notation in all model pages. Either:

• Convert the new model pages to use Credits/Token to match pricing-and-usage.md, or
• Add a note in each model page: "1 Credit ≈ $0.000001 USD" (or whatever the conversion rate is) so users can reconcile the two systems.

[MN-04] Auth Section Internally Inconsistent in API.md

Description

The Authentication section header labels the API Key method as "Messages endpoint only":

### API Key (Messages endpoint only)

However, the Messages endpoint description further down says:

**Auth:** API Key (`x-api-key`) or Bearer Token

If Bearer Token is now also accepted on /v1/messages, the authentication section header "Messages endpoint only" is misleading (it would imply API Key is only for Messages, and that Bearer is not supported there — the opposite of what is documented). The two statements need to be reconciled.

Recommendation

If both auth methods are now supported on /v1/messages, update the Authentication section:

### API Key

- **Type:** API Key
- **Header:** `x-api-key: <your-api-key>`
- **Endpoints:** `/v1/messages` (and `/v1/chat/completions` if applicable)

And update endpoint-level auth notes to be explicit:

**Auth:** Bearer Token or API Key (`x-api-key`)

[MN-05] Trailing Whitespace in Sidebar Configuration Files

Description

The newly added gpt-5-4-nano entry in both sidebar files has trailing whitespace. In sidebars.js there are 2 trailing spaces; in the i18n version there are 12. While JavaScript is not sensitive to this, it is inconsistent with the surrounding entries and may trigger linting rules (e.g., Prettier, ESLint no-trailing-spaces).

Code

// sidebars.js
'llmservice/models/gpt-5-4-nano',
// i18n/zh-Hans/.../sidebars.js
'llmservice/models/gpt-5-4-nano',

Recommendation

'llmservice/models/gpt-5-4-nano',

Suggestions

[S-01] Memory Service Docs Lack API/SDK Integration Section

File: docs/llmservice/memory.md
Description: The Memory service documentation exclusively covers user-facing concepts (auto-learning, import, incognito mode) and product advantages, but provides no API endpoint reference, SDK usage example, or parameter documentation for developers who want to programmatically interact with the Memory service.
Suggestion: Add a "Developer Integration" or "API Reference" section showing how to enable/disable memory via API parameters, how to retrieve or write memory nodes, or at minimum a link to the relevant API section if memory is toggled through a parameter on existing endpoints.

[S-02] Squash or Rebase Noisy Commit History Before Merge

File: Branch-level observation
Description: The PR has ~50 commits for what is effectively a documentation update. A significant number of commits represent rename-undo-rename cycles (e.g., chatgpt-5-4.md → gpt-5-4.md → chatgpt-5-4.md → gpt-5-4.md repeated across multiple files). This makes git blame and git bisect noisier for future maintainers.
Suggestion: Before merging, squash related commits into logical units (e.g., "Add GPT-5.4 model documentation", "Rename ChatGPT-5.x models to GPT-5.x", "Update pricing table with new models"). This is optional but improves maintainability.

Positive Observations

Checklist Results

Disclaimer

This is an automated code review. It supplements but does not replace human review. The reviewer analyzed only the diff between origin/main and origin/ai-bankofai-patch-1. Runtime behavior, integration testing, and deployment impact are not covered. All issues identified are within the documentation files changed by this PR; no unchanged files were audited.

Report generated by Code Review Skill v1.0.0
Date: 2026-04-06

Code Review Report

Project: BofAI Documentation (Docusaurus)
PR: main → ai-bankofai-patch-1
Review Date: 2026-04-06
Reviewer: AI Code Reviewer (Code Review Skill v1.0.0)

PR Overview

Branch Information

Commit History (Recent)

Review Summary

Verdict

Verdict: Request Changes

Findings at a Glance

Summary

This PR is a documentation-only update to a Docusaurus site. It introduces three categories of changes: (1) a significant overhaul of the LLM Service API reference — including a base URL domain migration, endpoint reordering, and richer inline examples; (2) model file renames (chatgpt-5-* → gpt-5-*) and three new GPT-5.4 family model pages; and (3) a brand-new Memory Service page plus updated pricing tables in both English and Simplified Chinese locales.

Overall quality is good: the API docs are substantially clearer and more complete in their new form, the new model pages are detailed and well-structured, and the bilingual Memory Service doc is consistent between locales. However, three issues require attention before merge. Most notably, the English pricing calculation example contains arithmetically incorrect rates for GPT-5.2 (the Chinese counterpart was fixed; the English one was not). The base URL domain migration (bankofai.com → bankofai.io) carries no migration notice, which could silently break existing integrations. And the Messages API example response uses "model": "gpt-5" instead of a Claude model ID, which directly contradicts the stated purpose of that endpoint and will confuse developers.

Change Summary

1. API Reference Overhaul

Purpose: Replace terse table-based schema documentation with richer, example-driven reference docs aligned with the new api.bankofai.io base URL.

2. Model File Renames

Purpose: Standardise model naming to drop the "ChatGPT" prefix.

3. New Model Pages

Purpose: Document the newly supported GPT-5.4 model family.

4. Memory Service Documentation

Purpose: Introduce user-facing documentation for the new persistent memory feature.

5. Pricing & Usage Updates

Purpose: Keep pricing documentation current with the expanded model catalogue and the TRON-only deposit constraint.

6. Sidebar Configuration

Purpose: Expose newly added pages in the navigation tree.

Detailed Findings

Major

[MJ-01] English Pricing Example Uses Wrong Rates for GPT-5.2

Description

The English pricing page was updated to reference GPT-5.2 in its worked calculation example, but the arithmetic still uses the old placeholder rates (1.25 input, 10.00 output) rather than GPT-5.2's actual rates (1.75 input, 14.00 output). The result, 512.5 credits, is therefore wrong for GPT-5.2.

The Chinese locale was correctly updated to use 1.75/14.00 and arrives at 717.5 credits — confirming the intent was to fix this example but the English version was missed.

Code (new English text):

For example, if you use GPT-5.2 to ask a question (10 input tokens) and the AI responds
with an answer (50 output tokens), the entire dialogue consumes 512.5 credits
(calculated as: 10 × 1.25 + 50 × 10).

Expected (matching Chinese locale and GPT-5.2 pricing):

For example, if you use GPT-5.2 to ask a question (10 input tokens) and the AI responds
with an answer (50 output tokens), the entire dialogue consumes 717.5 credits
(calculated as: 10 × 1.75 + 50 × 14.00).

Recommendation: Replace the rate values and result with the correct GPT-5.2 figures to match the Chinese locale and the pricing table on the same page.

[MJ-02] Base URL Domain Migration Has No Migration / Deprecation Notice

Description

The documented production base URL changes from https://api.bankofai.com to https://api.bankofai.io with no callout, changelog entry, or deprecation timeline. Developers who already have the .com URL hardcoded in their integrations will silently experience failures if the old domain is eventually decommissioned.

Code:

-| `https://api.bankofai.com` | Production |
+- **Base URL:** `https://api.bankofai.io`

Recommendation: Add an info/warning callout block immediately below the base URL noting the migration:

> **⚠️ Base URL Migration:** The production base URL has changed from
> `https://api.bankofai.com` to `https://api.bankofai.io`.
> The old domain will remain active until **[DATE]**. Please update
> your integrations before that date.

[MJ-03] Messages API Response Example Shows Wrong Model ID

Description

The Messages endpoint (POST /v1/messages) is described as Claude-compatible and its request example uses "model": "claude-sonnet-4-6". However, the non-stream response JSON example sets "model": "gpt-5", which is an OpenAI model ID. This is internally contradictory and will confuse developers who are using this section to understand what a real Claude response looks like.

Code:

{
  "id": "chatcmpl-xxx",
  "type": "message",
  "role": "assistant",
  "content": [
    { "type": "text", "text": "Hello! How can I help you?" }
  ],
  "stop_reason": "end_turn",
  "model": "gpt-5",           // ← should be a Claude model ID
  ...
}

Recommendation: Replace "gpt-5" with a Claude model identifier that matches the request example, e.g.:

  "model": "claude-sonnet-4-6",

Minor

[MN-01] H1 Headings Removed from Three Renamed Model Pages

Description

All three renamed model files had their # Title H1 heading removed and replaced with a blank line. The new GPT-5.4 model pages also start at ## Overview (H2) without an H1. Pages without H1 headings affect SEO, browser tab titles inferred from the document, and accessibility.

Code (diff from gpt-5-2.md):

-# ChatGPT-5.2
+

 ## Overview

Recommendation: Add an H1 heading to each file that matches the model's display name, consistent with the sidebar label:

# GPT-5.2

Apply the same fix to the three new GPT-5.4 model pages (gpt-5-4.md, gpt-5-4-mini.md, gpt-5-4-nano.md) which also lack H1 headings.

[MN-02] Page-Level H1 Heading Removed from Pricing Pages

Description

The # Pricing and Usage (EN) and # 计费与使用 (zh-Hans) H1 headings were removed from both pricing pages. Pages now open directly with ## Credits & Pricing (H2), leaving them without a top-level heading.

Code:

-# Pricing and Usage
-
 ## Credits & Pricing

Recommendation: Restore H1 headings:

# Pricing and Usage

## Credits & Pricing

[MN-03] GPT-5.4 Nano Pricing Shows Inconsistent Decimal Formatting

Description

The output price for GPT-5.4 Nano is listed as $1.2 in both EN and zh-Hans model files, while all other price entries in the codebase consistently use two decimal places (e.g. $1.20, $4.50). The main pricing table in pricing-and-usage.md correctly shows $1.20 for the same model.

Code:

| GPT-5.4 Nano |  $0.20 |  $1.2 |

Recommendation: Standardise to two decimal places:

| GPT-5.4 Nano |  $0.20 |  $1.20 |

[MN-04] Trailing Whitespace in Sidebar Configuration Files

Description

The gpt-5-4-nano sidebar entries in both sidebars.js files contain trailing whitespace after the trailing comma.

Code:

'llmservice/models/gpt-5-4-nano',

(two trailing spaces after the comma)

Recommendation: Remove trailing whitespace:

'llmservice/models/gpt-5-4-nano',

[MN-05] Deposit Section Silently Drops BNB Chain / BscScan Support

Description

The original documentation stated the platform supported "various mainstream tokens on supported networks (including TRON and BNB Chain)" and that users could verify transactions on "TRONSCAN or BscScan". The updated documentation narrows this to TRON-only (TRX, USDT, USDD, USD1) with no notice to existing BNB Chain depositors.

Code (old):

**Supported Token Types:** The platform supports various mainstream tokens on supported
networks (including TRON and BNB Chain).

Code (new):

**Supported Token Types:** The platform supports various mainstream tokens within the
TRON ecosystem, including TRX, USDT, USDD, and USD1.

Recommendation: If BNB Chain support has been discontinued, add a callout noting the change and any migration steps for users holding BNB Chain balances. If it is still supported but simply not mentioned, restore the reference.

[MN-06] Endpoint Reordering May Break Existing Deep Links

Description

The order of the API endpoints was changed. Previously: (1) Models, (2) Messages, (3) Chat Completions. After this PR: (1) Models, (2) Chat Completions, (3) Messages. The section heading anchors also changed (e.g. #messages → #3-messages-claude-compatible). Any existing bookmarks, external links, or internal cross-references pointing to old anchors will silently 404.

Recommendation: Either (a) keep the original anchor IDs by adding explicit Markdown anchor tags, or (b) document the changed section links in the PR description so that any external sites linking to the old anchors can be updated.

Suggestions

[S-01] Sidebar Model Ordering Could Follow Capability Tier

File: sidebars.js, i18n/zh-Hans/.../sidebars.js
Description: The current model list order is gpt-5-mini → gpt-5-nano → gpt-5-2 → gpt-5-4 → gpt-5-4-mini → gpt-5-4-nano. Placing nano before mini, and mixing GPT-5 generations in non-obvious order, may confuse users navigating the sidebar.
Suggestion: Order models from most capable to least within each generation, e.g.: gpt-5-4 → gpt-5-4-mini → gpt-5-4-nano → gpt-5-2 → gpt-5-mini → gpt-5-nano, then Claude models, then other providers.

[S-02] New Model Pages Missing Knowledge Cutoff Callout Box

File: docs/llmservice/models/gpt-5-4.md, gpt-5-4-mini.md, gpt-5-4-nano.md
Description: The knowledge cutoff date (August 31, 2025) is buried in the capabilities table. Other model pages in the codebase use a prominent callout or frontmatter for this.
Suggestion: Move knowledge cutoff information into a visible info/note callout near the top of each page so developers can see it at a glance before scrolling to the table.

[S-03] API Doc Schema Cross-References Removed Without Replacement

File: docs/llmservice/api/API.md
Description: The old API doc used formal named schema cross-references (e.g. [V1ModelsResponse](#v1modelsresponse), [ErrorResponse](#errorresponse)). These were removed in favour of inline JSON examples. The inline examples are easier to read, but formal schemas allowed machine-readable tooling (e.g. Postman, OpenAPI parsers) to understand the data model.
Suggestion: Consider adding an OpenAPI spec file (.yaml/.json) as the authoritative machine-readable contract, keeping the Markdown docs as the human-readable guide that derives from it.

Positive Observations

Checklist Results

Disclaimer

This is an automated code review. It supplements but does not replace human review. The reviewer analyzed only the diff between the specified branches. Runtime behavior, integration testing, and deployment impact are not covered.

Report generated by Code Review Skill v1.0.0
Date: 2026-04-06

Code Review Report

Project: BofAI Documentation Site
PR: main → ai-bankofai-patch-1
Review Date: 2026-04-06
Reviewer: AI Code Reviewer (Code Review Skill v1.0.0)

PR Overview

Branch Information

Commit History (Recent)

Review Summary

Verdict

Verdict: Request Changes

Findings at a Glance

Summary

This PR is a documentation-only update that expands the BofAI docs site in several meaningful ways: a complete API reference overhaul (both EN and ZH), new model pages for the GPT-5.4 family, a brand-new Memory service page, a rebranding of "ChatGPT-5" models to "GPT-5", and a payment-stack clarification (TRON-only). The overall direction is excellent — the API docs are dramatically more useful with inline JSON examples, full parameter tables, and stream event listings.

However, several issues require attention before merge. The most impactful is a widespread missing H1 heading across six model files (three renamed, three newly created), which affects page rendering, SEO, and Docusaurus navigation. A wrong model name in the Messages API response example ("model": "gpt-5" instead of a Claude model ID) will directly mislead developers integrating the Claude-compatible endpoint. There is also a pricing data inconsistency — the output price for GPT-5.4 Nano is shown as $1.2 in the model page but $1.25 in the pricing table — and an incomplete sentence left in the pricing page that references a link that was never inserted.

Change Summary

1. API Reference Overhaul (EN + ZH)

Purpose: Transform a sparse table-based API reference into a developer-friendly reference with inline code examples, explicit request/response schemas, and stream event documentation for both OpenAI-compatible and Claude-compatible endpoints.

2. Memory Service Documentation (New)

Purpose: Document the Memory (persistent context) service for users, covering smart learning, history import, user control, and incognito mode.

3. Model Rebranding: ChatGPT-5 → GPT-5

Purpose: Align public-facing model names with gpt-5.* naming convention, removing the chatgpt- prefix.

4. New GPT-5.4 Family Model Pages

Purpose: Document three newly released GPT-5.4-series models with overviews, benchmarks, use cases, capability tables, and pricing.

5. Pricing & Usage Page Updates

Purpose: Add pricing for MiniMax-M2.5, Kimi-K2.5, GLM-5, GPT-5.4 family; clarify TRON-only deposit; align example calculation to real model pricing.

Detailed Findings

Major

[MJ-01] Wrong Model Name in Messages API Response Example

Description

The non-streaming response example for the POST /v1/messages (Claude-compatible) endpoint shows "model": "gpt-5" in the returned JSON. However, this endpoint is designed for Claude models (e.g., claude-sonnet-4-6). A developer integrating the Claude-compatible endpoint would expect the response model field to reflect the Claude model ID they submitted, not "gpt-5". This is a direct factual error in a code example that developers may copy-paste or use to validate their integration.

Code

{
  "id": "chatcmpl-xxx",
  "type": "message",
  "role": "assistant",
  "content": [
    { "type": "text", "text": "Hello! How can I help you?" }
  ],
  "stop_reason": "end_turn",
  "model": "gpt-5",        <-- Should be a Claude model ID
  ...
}

Recommendation

{
  "id": "msg-xxx",
  "type": "message",
  "role": "assistant",
  "content": [
    { "type": "text", "text": "Hello! How can I help you?" }
  ],
  "stop_reason": "end_turn",
  "model": "claude-sonnet-4-6",
  ...
}

Additionally, the id prefix should likely be msg-xxx (Anthropic convention) rather than chatcmpl-xxx (OpenAI convention) to distinguish the two endpoints clearly. The same fix must be applied to the Chinese translation file.

[MJ-02] All Model Pages Missing H1 Heading

Description

All six English model files (and their six Chinese counterparts) lack an H1 (#) heading. The three renamed files had their H1 headings deleted (# ChatGPT-5.2, # ChatGPT-5-mini, # ChatGPT-5-nano) and replaced with a blank line. The three newly created files (gpt-5-4.md, gpt-5-4-mini.md, gpt-5-4-nano.md) were written starting directly at the ## Overview level.

In Docusaurus, a missing H1 means: (1) the page's <title> tag may fall back to the sidebar label, (2) og:title and SEO metadata are degraded, (3) heading hierarchy is broken (the first visible heading is ##), and (4) screen readers and accessibility tooling are impacted.

Code

-# ChatGPT-5.2
+

 ## Overview

(Similarly missing in all other model files)

Recommendation

Add an H1 heading matching the model's display name to each affected file. For example:

# GPT-5.2

## Overview
GPT-5.2 is the latest generation...

For new files:

# GPT-5.4

## Overview
GPT-5.4 is OpenAI's flagship...

[MJ-03] Pricing Data Inconsistency: GPT-5.4 Nano Output Price

Description

The body text of gpt-5-4-nano.md states the output price as $1.25/1M output tokens ("$0.20/1M input + $1.25/1M output"), but the pricing table at the bottom of the same file and in the main pricing table shows $1.2. This mismatch will confuse users trying to calculate costs.

Code

<!-- In the Key Features list -->
- **Extreme Cost Efficiency**: $0.20/1M input + $1.25/1M output ...

<!-- In the pricing table -->
| GPT-5.4 Nano | $0.20 | $1.2 |

Recommendation

Standardize to the correct value. If the price is $1.25, update the table:

| GPT-5.4 Nano | $0.20 | $1.25 |

If the price is $1.20, update the Key Features sentence. Apply the same fix to the Chinese translation.

[MJ-04] Base URL Changed Without Migration/Deprecation Notice

Description

The production server URL was silently changed from https://api.bankofai.com to https://api.bankofai.io. There is no deprecation notice, no migration guide, and no mention of when the old URL will stop working (if ever). Developers who integrated with the old URL and rely on these docs will not know they need to update their base URL. This is especially impactful for users who copied the URL directly from the old docs.

Code

-| `https://api.bankofai.com` | Production |
+- **Base URL:** `https://api.bankofai.io`

Recommendation

Add a callout or note near the Base URL clarifying the change and any transition period:

> **Note:** The production base URL has changed to `https://api.bankofai.io`.
> The previous URL (`https://api.bankofai.com`) will continue to work until [date].
> Please update your integration accordingly.

[MJ-05] Incomplete Sentence in Pricing Page ("please check here:")

Description

The pricing page ends a sentence with "please check here:" but provides no link or anchor reference. The text reads: "For specific model pricing, please check here:" and then immediately jumps to the pricing table. While the table follows, the sentence fragment is grammatically broken and implies a missing hyperlink (e.g., to an external page or a "Model Catalog" section).

Code

**Pricing for Different Models:** ...Some models do not support web search and are marked with "-".
For specific model pricing, please check here:

| Model | Input ...

Recommendation

Either remove the trailing phrase or complete it with a valid anchor/link:

**Pricing for Different Models:** ...Some models do not support web search and are marked with "-".
Refer to the table below for specific model pricing:

| Model | Input ...

Minor

[MN-01] H1 Heading Removed from Pricing and Usage Pages

Description

The PR removes the # Pricing and Usage H1 (and Chinese # 计费与使用) from both pricing files, causing the page to start with ## Credits & Pricing. Same SEO and accessibility concern as MJ-02, though less severe since the sidebar label "Pricing and Usage" will still appear as the rendered page title in Docusaurus.

Recommendation

Restore the H1 headings:

# Pricing and Usage

## Credits & Pricing

[MN-02] Authentication Section Inconsistency (x-api-key Scope)

Description

The Authentication section header now reads "API Key (Messages endpoint only)", implying x-api-key is exclusively for the Messages endpoint. However, the Messages endpoint description (POST /v1/messages) now says it accepts "API Key (x-api-key) or Bearer Token". These two statements are not contradictory, but the parenthetical "(Messages endpoint only)" next to the API Key section header is misleading — it sounds like Bearer Token cannot be used for Messages, when in fact both can.

Recommendation

Update the header to remove the ambiguity:

### API Key

- **Type:** API Key
- **Header:** `x-api-key: <your-api-key>`
- **Supported on:** Messages endpoint (`/v1/messages`)

Then in the endpoint section, clarify: "Auth: x-api-key (required) or Bearer Token (also accepted)".

[MN-03] Sidebar Navigation Order: GPT-5.4 Models Interleaved with GPT-5

Description

The new sidebar order is: gpt-5-mini, gpt-5-nano, gpt-5-2, gpt-5-4, gpt-5-4-mini, gpt-5-4-nano. This mixes the GPT-5 generation (mini, nano, 5.2) with the GPT-5.4 generation (5.4, 5.4-mini, 5.4-nano) in a non-intuitive way. The flagship gpt-5-2 appears before gpt-5-4 but after two "lite" variants, and the variant ordering within the GPT-5.4 family (standard → mini → nano, i.e., largest-to-smallest) differs from the GPT-5 family ordering (mini → nano → 5.2, i.e., smallest-to-largest then flagship).

Recommendation

Group by family and order consistently (flagship first, or smallest-to-largest — pick one):

// Option A: Flagship first within each family, newest family first
'llmservice/models/gpt-5-4',
'llmservice/models/gpt-5-4-mini',
'llmservice/models/gpt-5-4-nano',
'llmservice/models/gpt-5-2',
'llmservice/models/gpt-5-mini',
'llmservice/models/gpt-5-nano',

[MN-04] Trailing Whitespace in sidebars.js

Description

The gpt-5-4-nano sidebar entry has trailing whitespace (visible in the diff). While not functionally harmful in a JS config file, this is inconsistent with surrounding entries and should be cleaned up per standard code hygiene.

Code

'llmservice/models/gpt-5-4-nano',  ← trailing whitespace here

Recommendation

Remove trailing whitespace: 'llmservice/models/gpt-5-4-nano',

[MN-05] Deposit Section Silently Removes BNB Chain Support

Description

The old pricing docs stated the platform supports "various mainstream tokens on supported networks (including TRON and BNB Chain)". The new text narrowly scopes this to "TRON ecosystem" tokens only (TRX, USDT, USDD, USD1) and removes BNB Chain entirely. The Billing section also drops the reference to BscScan. Existing users who deposited via BNB Chain may be confused about whether their deposits are still visible or if the network is still supported. If BNB Chain has truly been dropped, existing users deserve an explicit notice.

Recommendation

Either add a deprecation callout for BNB Chain support, or briefly acknowledge legacy transaction history:

> **Note:** BNB Chain deposits are no longer supported for new top-ups as of [date].
> Existing BNB Chain transaction history remains viewable via BscScan.

Suggestions

[S-01] New GPT-5.4 Model Pages Should Include a "Release Date" Callout

File: docs/llmservice/models/gpt-5-4.md, gpt-5-4-mini.md, gpt-5-4-nano.md
Description: The release date (March 5, 2026 / March 17, 2026) is embedded in the Overview prose but not prominently surfaced. A dedicated callout or metadata field would help users quickly understand recency.
Suggestion: Add a metadata table or :::info callout at the top of each new model page showing "Released: March 17, 2026" and "Status: Generally Available".

[S-02] Memory Service Page Lacks API/Integration Details

File: docs/llmservice/memory.md
Description: The Memory page does a good job explaining the feature from a user perspective, but provides no API parameters, SDK examples, or integration guidance for developers who want to enable/disable memory programmatically or query memory nodes via API.
Suggestion: Add a "Developer Integration" section describing any API parameters (e.g., a memory_id field, a header to enable/disable memory per-request) and link to API reference where applicable.

[S-03] Pricing Example Should Use a Consistent Model Reference

File: docs/llmservice/pricing-and-usage.md and ZH equivalent
Description: The old pricing example used hypothetical rates (1.25/10.00) that didn't match any table row. The new example correctly uses GPT-5.2 (1.75/14.00) — this is a good improvement. However, to maximize clarity, consider adding the formula explicitly alongside the prose calculation.
Suggestion:

Total = (input_tokens × input_rate) + (output_tokens × output_rate)
      = (10 × 1.75) + (50 × 14.00)
      = 17.5 + 700
      = 717.5 credits

Positive Observations

Checklist Results

Disclaimer

This is an automated code review. It supplements but does not replace human review. The reviewer analyzed only the diff between the specified branches. Runtime behavior, integration testing, and deployment impact are not covered.

Report generated by Code Review Skill v1.0.0
Date: 2026-04-06

Code Review Report

Project: Bank of AI (BofAI) Documentation Site
PR: main → ai-bankofai-patch-1
Review Date: 2026-04-07
Reviewer: AI Code Reviewer (Code Review Skill v1.0.0)

PR Overview

Branch Information

Commit History (Recent, Representative)

Review Summary

Verdict

Verdict: ⚠️ Request Changes

Findings at a Glance

Summary

This PR is a documentation-only update to a Docusaurus-based documentation site for the Bank of AI (BofAI) LLM gateway service. It delivers several meaningful improvements: a comprehensive API reference overhaul (adding the Claude-compatible /v1/messages endpoint, full JSON examples, and richer data-model tables), a new Memory service documentation page, a batch of new GPT-5.4 family model pages, and a model naming migration from the chatgpt-5-* prefix to gpt-5-*. The overall quality of writing is good, structure is well-organized, and the i18n Chinese mirror tracks the English changes appropriately.

However, several issues require resolution before merge. The most impactful is that the newly added GPT-5.4 Pro model page is orphaned — it does not appear in either the English or Chinese sidebar navigation, making it unreachable through normal browsing. The pricing table on the pricing page is similarly incomplete, omitting GPT-5.4 Pro's rates. A pricing unit inconsistency between the per-model docs (which quote USD per 1M tokens) and the central pricing table (which quotes Credits per Token) will create serious user confusion. Finally, the Deposit section in the pricing page dropped its mention of BNB Chain support, which is a potential documentation regression if that deposit path is still active.

Change Summary

1. API Reference Overhaul (docs/llmservice/api/API.md + zh-Hans mirror)

Purpose: Significantly improves the API reference from terse table-only descriptions to full, developer-friendly documentation with concrete request/response examples, SSE event tables, and the complete Anthropic-compatible Messages endpoint.

2. New Memory Service Documentation (docs/llmservice/memory.md + zh-Hans mirror)

Purpose: Introduces a brand-new doc page explaining the platform's long-term memory/context persistence feature.

3. Model Renaming: chatgpt-5-* → gpt-5-*

Purpose: Aligns model page filenames/identifiers with the updated model ID convention.

4. New GPT-5.4 Family Model Pages

Purpose: Adds dedicated documentation pages for the new GPT-5.4 model family.

5. Pricing & Usage Page Update (docs/llmservice/pricing-and-usage.md + zh-Hans mirror)

Purpose: Reflects newly added models and updated deposit instructions.

6. Sidebar Navigation Updates (sidebars.js + zh-Hans mirror)

Purpose: Keeps navigation consistent with renamed and newly added model pages.

Detailed Findings

Major

[MJ-01] GPT-5.4 Pro Model Page Orphaned — Not in Sidebar Navigation

Description

The file docs/llmservice/models/gpt-5-4-pro.md (and its zh-Hans mirror) was added, but the model is never registered in either sidebars.js or the zh-Hans sidebar. As a result:

• The page is completely unreachable via normal sidebar navigation.
• Docusaurus may emit a build warning for an orphaned document.
• Users would only find this page through a direct URL or a manual search — undermining discoverability.

The three sibling models (gpt-5-4, gpt-5-4-mini, gpt-5-4-nano) were all correctly added to the sidebar. GPT-5.4 Pro was simply omitted.

Code

// sidebars.js — Models items list (current state, line ~193)
items: [
  'llmservice/models/gpt-5-mini',
  'llmservice/models/gpt-5-nano',
  'llmservice/models/gpt-5-2',
  'llmservice/models/gpt-5-4',
  'llmservice/models/gpt-5-4-mini',
  'llmservice/models/gpt-5-4-nano',   // <-- gpt-5-4-pro is MISSING here
  'llmservice/models/claude-haiku-4-5',
  ...

Recommendation

items: [
  'llmservice/models/gpt-5-mini',
  'llmservice/models/gpt-5-nano',
  'llmservice/models/gpt-5-2',
  'llmservice/models/gpt-5-4',
  'llmservice/models/gpt-5-4-mini',
  'llmservice/models/gpt-5-4-nano',
  'llmservice/models/gpt-5-4-pro',   // Add this entry
  'llmservice/models/claude-haiku-4-5',
  ...

Apply the same fix to i18n/zh-Hans/docusaurus-plugin-content-docs/current/sidebars.js.

[MJ-02] GPT-5.4 Pro Omitted from the Central Pricing Table

Description

The pricing-and-usage.md pricing table now lists GPT-5.4, GPT-5.4-mini, and GPT-5.4-nano, but GPT-5.4 Pro is absent. The model's own page (gpt-5-4-pro.md) quotes rates of 30.00 Credits/Token (input) and 180.00 Credits/Token (output) — the most expensive model in the catalogue by a wide margin. Users consulting the central pricing page cannot discover or budget for this model.

Code

<!-- pricing-and-usage.md — current table, GPT-5.4 Pro is nowhere present -->
| GPT-5.4           |                  2.50 |                  15.00 |                   10,000 |
| GPT-5.2           |                  1.75 |                  14.00 |                   10,000 |
| GPT-5.4-mini      |                  0.75 |                   4.50 |                   10,000 |
<!-- GPT-5.4 Pro MISSING -->

Recommendation

Add a row to the pricing table:

| GPT-5.4 Pro       |                 30.00 |                 180.00 |                   10,000 |

Apply the same fix to the zh-Hans mirror.

[MJ-03] Pricing Unit Inconsistency Between Model Pages and the Pricing Table

Description

The new model-specific pages quote pricing in USD per 1 million tokens:

<!-- gpt-5-4.md -->
| GPT-5.4 | $2.50 | $15.00 |   ← "per 1M tokens (USD)"

But the central pricing table quotes pricing in Credits per individual token:

<!-- pricing-and-usage.md -->
| GPT-5.4  | 2.50 | 15.00 |   ← "Credits/Token"

The numbers happen to be identical (2.50 and 15.00), which makes the discrepancy invisible at a glance but deeply misleading: $2.50 per 1M tokens ≠ 2.50 Credits per token. A user comparing the two pages will either assume they are the same unit (incorrect) or be thoroughly confused about actual costs. For GPT-5.4 Pro the model page says 30.00/180.00 — in dollar-per-million-tokens terms these are consistent, but a developer reading the pricing table would interpret them as Credits-per-token, implying a 12,000× cost difference versus the model page.

Code

<!-- gpt-5-4.md — header row says "per 1M tokens" in USD -->
| Model | Input (per 1M tokens) | Output (per 1M tokens) |
| :--- | :--- | :--- |
| GPT-5.4 | $2.50 | $15.00 |

<!-- pricing-and-usage.md — header row says "Credits/Token" -->
| Model | Input (Credits/Token) | Output (Credits/Token) |
| GPT-5.4 |  2.50 | 15.00 |

Recommendation

Decide on a single canonical unit (e.g., Credits per 1M tokens or Credits per token) and apply it consistently across all model pages and the pricing table. Update the column headers and values uniformly. Clarify the Credits-to-USD conversion rate (if any) in the pricing page introduction so users can cross-reference.

[MJ-04] Deposit Section Drops BNB Chain Support — Potential Documentation Regression

Description

The main branch version of the deposit section stated:

"The platform supports various mainstream tokens on supported networks (including TRON and BNB Chain)."

The new version now reads:

"The platform supports various mainstream tokens within the TRON ecosystem, including TRX, USDT, USDD, and USD1."

BNB Chain (and any associated tokens) is completely removed. If BNB Chain deposits remain available in the actual product, this change constitutes a documentation regression that will mislead users into believing only TRON is supported — potentially causing failed deposit attempts. The zh-Hans mirror carries the same TRON-only text.

Code

<!-- main branch -->
* **Supported Token Types:** The platform supports various mainstream tokens
  on supported networks (including TRON and BNB Chain).

<!-- ai-bankofai-patch-1 -->
**Supported Token Types:** The platform supports various mainstream tokens within
the TRON ecosystem, including TRX, USDT, USDD, and USD1.

Recommendation

If BNB Chain deposits are still supported, restore the multi-network language and enumerate supported tokens per network. If BNB Chain has been discontinued, the change is intentional but should be accompanied by a clear notice (e.g., a deprecation callout block) for existing users. Verify the product state before merging.

Minor

[MN-01] Missing H1 Title on Pricing Page

Description

The original file opened with a top-level heading # Pricing and Usage. The new version omits this and starts directly at an H2 (## Credits & Pricing). Docusaurus uses the first H1 as the browser tab title and breadcrumb label. Without it, the page title may fall back to the sidebar label or be blank, degrading the user experience.

Recommendation

Restore the H1 title:

# Pricing and Usage

## Credits & Pricing
...

Apply the same fix to the zh-Hans mirror.

[MN-02] Internal Cache Field Names Exposed in Messages API Response Example

Description

The non-streaming response example for the /v1/messages endpoint includes two unusual fields:

"claude_cache_creation_5_m_tokens": 0,
"claude_cache_creation_1_h_tokens": 0

These field names look like internal implementation artefacts (referencing raw Anthropic cache TTL durations). Exposing them as first-class documented API fields may:

1. Confuse developers trying to understand their meaning (5_m = 5 minutes? 1_h = 1 hour?).
2. Create implicit API contract obligations if clients start depending on them.
3. Leak internal infrastructure terminology.

Recommendation

Either remove these fields from the public-facing example (if they are internal), or rename and document them explicitly (e.g., cache_creation_tokens_5min, cache_creation_tokens_1hr) with a clear description of what each represents and when it is non-zero.

[MN-03] Trailing Whitespace in Sidebar Files

Description

Both sidebar files contain a line with trailing whitespace:

'llmservice/models/gpt-5-4-nano',

While harmless at runtime, trailing whitespace commonly triggers linter warnings, pollutes diffs, and is inconsistent with the rest of the file.

Recommendation

Remove the trailing spaces from both files:

'llmservice/models/gpt-5-4-nano',

[MN-04] API Base URL Changed Without Explicit Callout

Description

The API base URL was silently changed from https://api.bankofai.com (listed in the former Servers table) to https://api.bankofai.io (new header field). This change is significant for any developer who has already integrated against the documented endpoint. There is no changelog note, migration callout, or deprecation notice explaining the change. If the old domain is no longer the canonical API endpoint, existing integrations will break without warning.

Code

<!-- main branch -->
| `https://api.bankofai.com` | Production |

<!-- ai-bankofai-patch-1 -->
- **Base URL:** `https://api.bankofai.io`

Recommendation

Add an info/warning callout block in the API doc clarifying the domain change:

:::info Base URL Update
The production base URL has changed from `api.bankofai.com` to `api.bankofai.io`.
Please update any existing integrations accordingly.
:::

Additionally, verify that api.bankofai.com is either redirecting to api.bankofai.io or has been decommissioned before finalizing this documentation.

Suggestions

[S-01] GPT-5.4 Nano Output Pricing Formatted Inconsistently

File: docs/llmservice/models/gpt-5-4-nano.md
Description: The Credits and Pricing table in the Nano model page uses $1.2 for output, while all other model pages use two-decimal formatting ($1.20, $4.50, $15.00, etc.).
Suggestion: Normalise to two decimal places: change $1.2 → $1.20 for consistency.

[S-02] Duplicate Commit Pairs Suggest Force-Push History Was Not Cleaned

File: Git history
Description: The commit log contains several duplicate pairs with identical messages: "Create gpt-5-4-pro.md" appears twice (5328cda, 0044127), "Update pricing-and-usage.md" appears multiple times (2a6a7fd, a2b2d5f, 9e2e104, 8455399), and so on. This indicates iterative direct-to-branch commits without squashing, which creates noisy history.
Suggestion: Before merging, squash or consolidate related commits into logical units (e.g., "Add GPT-5.4 family model documentation", "Update API reference to v1.1") to produce a cleaner, reviewable commit history.

Positive Observations

Checklist Results

Disclaimer

This is an automated code review. It supplements but does not replace human review. The reviewer analyzed only the diff between origin/main and origin/ai-bankofai-patch-1. Runtime behavior, integration testing, and deployment impact are not covered.

Report generated by Code Review Skill v1.0.0
Date: 2026-04-07

Code Review Report

Project: Bank of AI Documentation (@x402-tron/docs)
PR: main → ai-bankofai-patch-1
Review Date: 2026-04-08
Reviewer: AI Code Reviewer (Code Review Skill v1.0.0)

PR Overview

Branch Information

Commit History (Recent)

Review Summary

Verdict

Verdict: ⚠️ Request Changes

Findings at a Glance

Summary

This PR is a significant documentation update that renames the GPT model family (ChatGPT-5.x → GPT-5.x), adds four new model pages (GPT-5.4, GPT-5.4 Mini, GPT-5.4 Nano, GPT-5.4 Pro), introduces a new Memory Service page, revamps the API reference with richer inline examples, and updates the pricing table. Both English (docs/) and Chinese (i18n/zh-Hans/) documentation are updated in parallel.

Overall the changes are well-structured and the quality of the new content is good. However, several factual inconsistencies and a breaking change require resolution before merge. The most impactful issues are: a silent change of the production API base URL with no migration guidance, a pricing discrepancy in GPT-5.4 Nano, a wrong model name in a response example, GPT-5.4 Pro absent from the pricing table, and the undocumented removal of BNB Chain deposit support. Additionally, seven documentation pages were left without H1 titles, which will render as untitled pages on the site.

Change Summary

1. Model Renames: ChatGPT-5.x → GPT-5.x

Purpose: Align model names with upstream OpenAI branding (dropping the "Chat" prefix).

2. New Model Documentation (GPT-5.4 Family)

Purpose: Document four newly supported models in the GPT-5.4 series.

3. API Reference Overhaul

Purpose: Improve readability, add concrete request/response examples, document the Claude-compatible Messages endpoint more thoroughly.

4. New Memory Service Page

Purpose: Introduce long-term memory capabilities for AI agents.

5. Pricing, Sidebar & Package Updates

Purpose: Surface new models and Memory in navigation; reflect deposit network changes; bump package version.

Detailed Findings

Major

[MJ-01] Breaking API Base URL Change with No Migration Guidance

Description

The production API base URL changed from https://api.bankofai.com to https://api.bankofai.io. This is a breaking change for any existing users who have hard-coded the old domain. The PR removes the old URL entirely and substitutes the new one with zero context: no deprecation notice, no redirect information, no migration timeline, and no changelog entry.

If the domain change is live on the backend, users relying on api.bankofai.com will receive connection errors with no guidance. If it is not yet live, the documentation is ahead of the infrastructure.

Code

-## Servers
-| URL | Description |
-|-----|-------------|
-| `https://api.bankofai.com` | Production |
+- **Base URL:** `https://api.bankofai.io`

Recommendation

Add an explicit deprecation notice near the base URL change. For example:

> **Note (Migration):** The API base URL has changed from `https://api.bankofai.com`
> to `https://api.bankofai.io`. The old domain will continue to accept traffic until
> **[DATE]**. Please update your integrations accordingly.

If the old domain is being kept as an alias indefinitely, document that. If it is being decommissioned, include the timeline.

[MJ-02] GPT-5.4 Nano Output Pricing Inconsistency ($1.25 vs $1.20)

Description

The Overview section (line 1) and the Key Features bullet state the output price as $1.25 per million tokens. However, the Credits and Pricing table at the bottom of the same file shows $1.2 (missing the trailing 5). Both the English and Chinese files share this inconsistency. Users relying on the pricing table — the most likely reference point — will see a lower figure than the one stated in the prose.

Code

# Overview (line 1):
"At just $0.20/$1.25 per MTok..."

# Key Features bullet:
"$0.20/1M input + $1.25/1M output"

# Credits and Pricing table:
| GPT-5.4 Nano | $0.20 | $1.2 |   ← should be $1.25

Recommendation

Decide on the canonical price and apply it consistently. If the price is $1.25:

| GPT-5.4 Nano | $0.20 | $1.25 |

Also verify consistency against the central pricing-and-usage.md table, which currently shows 1.20 credits/token for GPT-5.4-nano — all three locations must agree.

[MJ-03] Messages API Response Example Hardcodes Wrong Model Name

Description

The Messages endpoint (POST /v1/messages) request example sends "model": "claude-sonnet-4-6", which is correct. However, the corresponding non-stream response example returns "model": "gpt-5" — a completely different model family. This is factually wrong and will confuse developers building against the Claude-compatible endpoint: the response model field should echo back the model that was used ("claude-sonnet-4-6" in this case).

Code

// Request:
{ "model": "claude-sonnet-4-6", ... }

// Response (non-stream):
{
  "model": "gpt-5",   ← WRONG — should match the requested model
  ...
}

Recommendation

Update the response example to reflect the model sent in the request:

{
  "id": "msg-xxx",
  "type": "message",
  "role": "assistant",
  "model": "claude-sonnet-4-6",
  ...
}

[MJ-04] GPT-5.4 Pro Not Listed in Pricing Table

Description

Four new models are added (GPT-5.4, GPT-5.4 Mini, GPT-5.4 Nano, GPT-5.4 Pro) and the first three appear in the updated pricing table in pricing-and-usage.md. GPT-5.4 Pro is entirely absent from the central pricing table despite having its own model page that lists its credits (30.00 input / 180.00 output credits/token — the most expensive model in the lineup). Users comparing models on the pricing page will have no visibility into GPT-5.4 Pro's cost.

Code

# pricing-and-usage.md pricing table (excerpt):
| GPT-5.4           | 2.50 | 15.00 | 10,000 |
| GPT-5.2           | 1.75 | 14.00 | 10,000 |
| GPT-5.4-mini      | 0.75 |  4.50 | 10,000 |
# ... GPT-5.4-Pro is missing entirely

Recommendation

Add GPT-5.4 Pro to the pricing table:

| GPT-5.4 Pro       | 30.00 | 180.00 | 10,000 |

Also confirm whether GPT-5.4 Pro supports web search (the model page does not mention it) and set the web search column appropriately.

[MJ-05] Deposit Section Silently Removes BNB Chain Support

Description

The previous documentation explicitly listed BNB Chain as a supported deposit network alongside TRON, and referenced BscScan as a blockchain explorer for verifying BNB Chain transactions. This PR rewrites the Deposit section to mention only the TRON ecosystem (TRX, USDT, USDD, USD1) and removes any reference to BNB Chain or BscScan. There is no deprecation notice, no user-facing announcement, and no alternative offered for users who deposited or planned to deposit via BNB Chain.

If BNB Chain support has genuinely been removed, this is a significant breaking change for existing users with no communication. If it still exists but was accidentally omitted, this is a documentation error.

Code

-* **Supported Token Types:** The platform supports various mainstream tokens on
-  supported networks (including TRON and BNB Chain).
-* ... You can click the transaction hash to verify details on the corresponding
-  blockchain explorer (e.g., TRONSCAN or BscScan).
+**Supported Token Types:** The platform supports various mainstream tokens within
+ the TRON ecosystem, including TRX, USDT, USDD, and USD1.
+**Transaction Hash Query:** You can click the transaction hash to verify the
+ transaction details...

Recommendation

If BNB Chain is being deprecated:

• Add a notice in the Deposit section explaining the deprecation and effective date.
• Advise BNB Chain users of alternative steps.

If BNB Chain is still supported but accidentally omitted:

• Restore the BNB Chain references and BscScan explorer link.

Minor

[MN-01] Seven Model Pages and One Doc Page Missing H1 Title

Description

The three renamed model files had their # ChatGPT-5.x H1 headings removed without replacement:

-# ChatGPT-5.2
+
 ## Overview

Similarly, the four new model files (gpt-5-4.md, gpt-5-4-mini.md, gpt-5-4-nano.md, gpt-5-4-pro.md) and the updated pricing-and-usage.md all start with ## (H2) headings rather than an # (H1) title. In Docusaurus, a missing H1 means the page will render without a visible page title, relying solely on the sidebar label.

Recommendation

Add an appropriate H1 heading as the first line of each affected file:

# GPT-5.2
## Overview
...

# GPT-5.4 Pro
## Overview
...

# Pricing and Usage
## Credits & Pricing
...

[MN-02] GPT-5.4 Context Window Contradicts Itself

Description

The Overview paragraph says GPT-5.4 supports "up to 1,050,000 tokens of context", but the Capabilities and Limitations table gives two contradictory values:

# Overview (line 3):
"With up to 1,050,000 tokens of context..."

# Capabilities table:
| **Context Window** | Standard 272K tokens; expandable to 1,050,000 tokens (requires explicit configuration). |

The Overview implies 1M+ is the default; the table says 272K is the standard and 1M+ requires configuration. These should be reconciled so users know the default context size and how to unlock the larger window.

Recommendation

Decide on the canonical description. If 272K is the default:

# Overview: "... supports up to 272,000 tokens by default, expandable to 1,050,000 tokens."
# Table: unchanged (272K standard; 1,050K with explicit config)

[MN-03] GPT-5.4 Pro Has a Duplicate "Limitations" Section

Description

Limitations are documented twice in gpt-5-4-pro.md. The Capabilities and Limitations table already includes a Response Speed row noting high latency and a Cost consideration. The file then adds a separate ## Limitations section at the bottom repeating the same three points (high latency, cost, no audio/video input). This creates redundancy and a risk of the two sections diverging over time.

Code

# In the Capabilities table:
| **Response Speed** | Slower — some complex requests may take several minutes... |

# Then separately at the bottom:
## Limitations
- **High latency**: Complex requests may take several minutes...
- **Cost**: It is one of the most expensive API models available.
- **No audio/video input**: ...

Recommendation

Remove the standalone ## Limitations section and ensure all limitation details are captured in the Capabilities and Limitations table, consistent with the structure of other model pages (gpt-5-4.md, gpt-5-4-mini.md, etc.).

[MN-04] API.md Removes Several Documented Response Fields

Description

The old API.md contained detailed schema tables for ChatCompletionsResponse that documented fields including service_tier, system_fingerprint, and obfuscation. These fields are absent from the new inline JSON examples. While the examples are cleaner, developers who use the obfuscation or service_tier fields (e.g., for debugging or compliance) have lost their documentation reference.

Recommendation

Either restore these fields in the inline response examples with a comment noting they are optional/informational:

{
  "id": "chatcmpl-xxx",
  "service_tier": "default",
  "system_fingerprint": null,
  "obfuscation": "...",
  ...
}

Or add a note such as: "Additional response fields (service_tier, system_fingerprint, obfuscation) may appear; see full schema reference."

[MN-05] API Key Docs Describe Messages Endpoint as "Only" then Change Scope

Description

The Authentication section header says "API Key (Messages endpoint only)", implying x-api-key is restricted to Messages. However, the Messages endpoint description below states it accepts "API Key (x-api-key) or Bearer Token" — meaning Bearer Token also works on Messages. The "only" qualifier in the section header is therefore misleading; it should clarify that x-api-key is available on Messages, not that Messages exclusively uses it.

Code

### API Key (Messages endpoint only)   ← misleading — Bearer also works on Messages
...

### 3. Messages (Claude Compatible)
**Auth:** API Key (`x-api-key`) or Bearer Token   ← both auth methods accepted

Recommendation

Rename the auth section to clarify:

### API Key
- **Supported on:** `/v1/messages` endpoint
- **Type:** API Key
- **Header:** `x-api-key: <your-api-key>`

Suggestions

[S-01] "Zero-Hallucination Recall" Claim May Overpromise

File: docs/llmservice/memory.md
Description: The heading "Zero-Hallucination Recall" is a very strong claim. While the rationale (exact user-curated memory nodes vs. semantic approximation) is sound, no memory system can truly guarantee zero hallucination. This phrasing may set user expectations that cannot always be met and could expose the platform to credibility risk if the feature underdelivers.
Suggestion: Soften to something like "High-Fidelity Recall" or "Reduced Hallucination Risk" with a brief explanation of the mechanism, e.g.: "By grounding responses in exact, user-curated memory nodes, the system significantly reduces the risk of the AI misremembering or conflating past context."

[S-02] No Changelog or Migration Notice for Model Renames

File: docs/llmservice/pricing-and-usage.md, sidebars.js
Description: Renaming ChatGPT-5.x → GPT-5.x changes the model IDs users may have seen referenced in documentation. While the API model ID strings (e.g., "gpt-5.2") appear unchanged, the rename may confuse users who bookmarked old page URLs or sidebar links.
Suggestion: Add a brief note near the top of the renamed model pages (or in a dedicated changelog) such as: "Previously known as ChatGPT-5.2. The model identifier in API requests remains gpt-5.2."

[S-03] New Model Files Could Cross-Reference the Pricing Page

File: docs/llmservice/models/gpt-5-4-pro.md, gpt-5-4.md, gpt-5-4-mini.md, gpt-5-4-nano.md
Description: Each new model page includes a local pricing table. As the platform grows, keeping per-model pricing tables in sync with the central pricing-and-usage.md table will become a maintenance burden. The older model pages (e.g., gpt-5-2.md) already follow this pattern.
Suggestion: Consider replacing or supplementing the per-page pricing table with a link: "For up-to-date pricing, see the Pricing and Usage page." This creates a single source of truth for pricing.

Positive Observations

Checklist Results

Disclaimer

This is an automated code review. It supplements but does not replace human review. The reviewer analyzed only the diff between the specified branches. Runtime behavior, integration testing, and deployment impact are not covered.

Report generated by Code Review Skill v1.0.0
Date: 2026-04-08