Code Review Report

Project: Bank of AI — Documentation Site
PR: main -> ai-bankofai-patch-1
Review Date: 2026-04-27
Reviewer: AI Code Reviewer (Code Review Skill v1.0.0)

PR Overview

Branch Information

Commit History

Review Summary

Verdict

Verdict: Request Changes

Findings at a Glance

Summary

This PR adds documentation for three newly launched AI models — DeepSeek V4 Flash, DeepSeek V4 Pro, and GPT-5.5 — to the Bank of AI LLM Service documentation site. Each model receives a new English page, a Chinese (zh-Hans) translation, an entry in the master pricing table, and a sidebar registration in sidebars.js. The change is purely additive (no existing content is deleted or modified beyond the pricing table rows) and carries no security or performance risk.

Overall content quality is solid: the new pages are well-structured, use consistent Markdown formatting within themselves, and correctly parallel the site's established page layout. The zh-Hans translations appear accurate and complete.

One Major issue requires attention before merge: GPT-5.5 has important tiered pricing beyond 272K input tokens (2× rate) that is correctly documented on the model-specific page but is absent from the pricing-and-usage.md master pricing table, where the Notes column shows only "-". This omission can mislead users who rely on the summary table for cost planning. Additionally, five Minor issues cover cross-page inconsistencies around pricing units, missing knowledge-cutoff fields, a translation omission, absent source references for DeepSeek pages, and references to a model version (Claude Opus 4.7) not yet listed on the platform.

Change Summary

1. New English Model Pages (3 new files)

Purpose: Surface newly released (April 2026) model documentation so platform users can evaluate model capabilities and pricing before selecting a model.

2. Chinese (zh-Hans) Translations (3 new + 1 modified files)

Purpose: Keep English and Chinese documentation in parity for the three new models.

3. Pricing Table & Sidebar Updates (2 modified files)

Purpose: Make the new pages discoverable via site navigation and surfaced in the pricing reference table.

Detailed Findings

Major

[MJ-01] GPT-5.5 Tiered Pricing Note Missing from Master Pricing Table

Description

GPT-5.5 has a documented tiered pricing structure: input tokens beyond 272K are charged at 2× the base rate, and output tokens at 1.5×. This cost modifier is correctly documented on the model-specific page (docs/llmservice/models/gpt-5-5.md, pricing table Notes column), but the master pricing-and-usage.md table records "-" in the Notes column, giving no indication of the surcharge. Users who rely on the summary table for cost estimates could significantly underestimate the cost of long-context GPT-5.5 workloads.

Code

# pricing-and-usage.md (master table — what was added)
+| GPT-5.5           |                  5.00 |                  30.00 |                        - |

# gpt-5-5.md (model page — correctly documents the surcharge)
| **GPT-5.5** | `5.00` | `30.00` | 2x input / 1.5x output beyond 272K tokens |

Recommendation

Update the Notes column in the master pricing table to surface the tiered rate, consistent with the model page:

| GPT-5.5           |                  5.00 |                  30.00 |   2× input beyond 272K tokens |

The same correction should be applied to the zh-Hans pricing table:

| GPT-5.5 | 5.00 | 30.00 | 超过 272K 输入按 2 倍计费 |

Minor

[MN-01] Inconsistent Pricing Units Across Model-Specific Pages

Description

The new model pages use "Credits/Token" as the pricing unit in their individual tables. Existing model pages use different units: deepseek-v3.2.md uses "Credits/1K Tokens" with an explicit calculation note, and gpt-5-4.md uses "$ per 1M tokens". While the new pages correctly match the pricing-and-usage.md master table format (Credits/Token), a user reading multiple model pages side-by-side will encounter three different unit conventions.

Code

# New docs (Credits/Token) — correct match to pricing-and-usage.md
| **DeepSeek V4 Flash** | `0.14` | `0.28` | 1M context, 384K max output |

# deepseek-v3.2.md (Credits/1K Tokens) — older convention
| **DeepSeek-V3.2** | `0.27` | `0.42` |   ← table header: Credits/1K Tokens

# gpt-5-4.md ($ per 1M tokens) — yet another convention
| GPT-5.4 | $2.50 | $15.00 |   ← table header: per 1M tokens

Recommendation

Standardize all model-specific pricing tables to use "Credits/Token" (matching pricing-and-usage.md) in a follow-up cleanup PR targeting the older model pages. No change is needed in this PR — the new pages are the correct standard.

[MN-02] Knowledge Cutoff Date Missing from New Model Capability Tables

Description

The existing gpt-5-4.md capabilities table includes a "Knowledge Cutoff" row (August 31, 2025). None of the three new model pages include this field, leaving users without information about training data recency — a commonly asked question when evaluating models for knowledge-intensive tasks.

Code

# gpt-5-4.md — has this row (existing standard)
| **Knowledge Cutoff** | August 31, 2025 |

# gpt-5-5.md — missing this row entirely
| **Tool Use** | Full function calling, tool search, ... |
| **Multilingual** | Broad multilingual support ...          |
# No Knowledge Cutoff row

Recommendation

Add a Knowledge Cutoff row to each new model's capability table. Example for GPT-5.5:

| **Knowledge Cutoff** | April 2026 (verify with OpenAI announcement) |

[MN-03] DeepSeek Multilingual Description Drops "Chinese" in zh-Hans Translation

Description

The English deepseek-v4-flash.md states the model has "Broad multilingual support; strongest in English and Chinese". The zh-Hans translation renders this as "广泛支持多语言，其中英文表现最强" — which omits the mention of Chinese being among the strongest languages. While perhaps intentional (obvious to Chinese readers), it creates a factual discrepancy between the two language versions.

Code

# English (docs/llmservice/models/deepseek-v4-flash.md)
| **Multilingual** | Broad multilingual support; strongest in English and Chinese |

# zh-Hans translation — "Chinese" omitted
| **多语言能力** | 广泛支持多语言，其中英文表现最强 |

Recommendation

Update the zh-Hans translation to include both languages for accuracy:

| **多语言能力** | 广泛支持多语言，英文与中文表现最强 |

[MN-04] DeepSeek Model Pages Lack Source References

Description

The gpt-5-5.md page includes a References section with links to the official OpenAI announcement and pricing page, which is good practice for verifying claimed benchmark scores. The two new DeepSeek pages cite specific benchmark numbers (SWE-bench, LiveCodeBench, Codeforces rating, MCPAtlas, etc.) but provide no references, making it impossible for users to verify the claims or read the source material.

Code

# gpt-5-5.md — has references (good practice)
## References
[1] [OpenAI GPT-5.5 Announcement](https://openai.com/index/gpt-5-5)
[2] [OpenAI API Pricing](https://openai.com/api/pricing)

# deepseek-v4-pro.md — ends without references
| **DeepSeek V4 Pro** | `1.74` | `3.48` | 1M context, 384K max output |
# EOF — no references section

Recommendation

Add a References section to both DeepSeek pages linking to the official DeepSeek release blog or GitHub repo (e.g., https://github.com/deepseek-ai/DeepSeek-V4).

[MN-05] References to "Claude Opus 4.7" — Model Not Listed on the Platform

Description

Two new model pages reference Claude Opus 4.7 in comparisons, but the platform's model catalog (and sidebar) only lists up to Claude Opus 4.6. This may cause user confusion — readers may attempt to find Claude Opus 4.7 on the platform, or question the accuracy of the entire comparison table.

Code

# deepseek-v4-pro.md — Line 7
...roughly 1/20th the price of Claude Opus 4.7.

# gpt-5-5.md — Known Limitations, Line ~45
Lost the harder SWE-Bench Pro benchmark to Claude Opus 4.7 despite winning the
standard SWE-Bench Verified headline.

Recommendation

Either:

1. Replace "Claude Opus 4.7" with the correct currently available model (Claude Opus 4.6) if this is a documentation error, or
2. Add a note clarifying that Claude Opus 4.7 is the anticipated successor and not yet available on the platform.

Suggestions

[S-01] DeepSeek Model Pages Could Adopt Section Emoji Style Matching deepseek-v3.2.md

File: docs/llmservice/models/deepseek-v4-flash.md, docs/llmservice/models/deepseek-v4-pro.md
Description: The existing deepseek-v3.2.md uses emoji prefixes on section headings (🚀 Key Features, 💡 Best Use Cases, 📊 Capabilities, 💰 Credits). The new DeepSeek pages use plain headings without emoji, creating slight visual inconsistency within the DeepSeek model family pages.
Suggestion: Adopt emoji headers to match deepseek-v3.2.md style, or update deepseek-v3.2.md in a cleanup PR to match the plainer style. Either way, standardize within the DeepSeek family.

[S-02] Explicitly Document Web Search Support Status in Model Capability Tables

File: docs/llmservice/models/gpt-5-5.md, docs/llmservice/models/deepseek-v4-flash.md, docs/llmservice/models/deepseek-v4-pro.md
Description: The master pricing table has a "Web Search (Credits/Use)" column that shows - for all three new models, meaning web search is unsupported. However, the model-specific capability tables do not include a row for web search availability. Users reading only the model page may not know this feature is unavailable.
Suggestion: Add a "Web Search" row to the capability table with "Not supported" for all three new models. This small addition prevents user confusion and avoids support tickets.

[S-03] Add a "Release Date" Row to Capability Tables

File: All three new model pages
Description: The new model docs describe very recent models (released April 23–24, 2026) but don't include a formal "Release Date" field in the capabilities table. As the model catalog grows, release date becomes useful for understanding the recency of a model relative to alternatives.
Suggestion: Add a "Release Date" row to each capabilities table. Example: | **Release Date** | April 24, 2026 |

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 (main...ai-bankofai-patch-1). Benchmark accuracy, pricing correctness relative to upstream providers, and live site rendering were not independently verified.

Report generated by Code Review Skill v1.0.0
Date: 2026-04-27

Code Review Report

Project: x402-tron / BANK OF AI Docs
PR: main → ai-bankofai-patch-1
Review Date: 2026-04-28
Reviewer: AI Code Reviewer (Code Review Skill v1.0.0)

PR Overview

Branch Information

Commit History

Review Summary

Verdict

Verdict: Request Changes

Findings at a Glance

Summary

This PR adds documentation pages for three newly-launched LLM models — GPT-5.5, DeepSeek V4 Pro, and DeepSeek V4 Flash — across both the English and Chinese (zh-Hans) locales. It registers each model in the navigation sidebars, updates the global pricing-and-usage table, and bumps the package version from 1.2.12 to 1.2.13. The documentation content is accurate, well-structured, and consistent in benchmark figures across locales.

One Major issue requires a fix before merging: the global pricing page omits GPT-5.5's long-context tiered surcharge (2x input / 1.5x output beyond 272K tokens), which is a significant billing detail that users relying on the pricing table alone will miss. Three Minor issues cover a missing H1 title in the GPT-5.5 document (both locales), a numerical inconsistency in the GPT-5.5 overview text, and a vague commit message. Four Suggestions address consistency improvements and user-experience enhancements.

Change Summary

1. New Model Documentation — English (docs/llmservice/models/)

Purpose: Expose newly-released model options to English-speaking users.

2. New Model Documentation — Chinese (i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/models/)

Purpose: Localise new model documentation for Chinese-speaking users.

3. Pricing Table Updates

Purpose: Keep the central pricing reference up to date.

4. Sidebar Navigation Registration

Purpose: Make new model pages discoverable via the site navigation.

5. Package Version Bump

Purpose: Reflect the documentation content addition in the package version.

Detailed Findings

Major

[MJ-01] GPT-5.5 Long-Context Surcharge Not Surfaced in Global Pricing Table

Description

GPT-5.5 has a tiered pricing model: input tokens beyond 272K are billed at 2x and output tokens beyond 272K at 1.5x the base rate. This billing caveat is documented in docs/llmservice/models/gpt-5-5.md (pricing table Notes column and Known Limitations) and in the Chinese model page. However, the global pricing reference (pricing-and-usage.md) — the page most users will consult first to estimate costs — has no Notes column and therefore completely omits this surcharge. A user who reads only the pricing table will see 5.00 / 30.00 Credits/Token and have no indication that long-context requests can cost double. This could lead to significant unexpected charges for users running large-context workloads.

Code

<!-- docs/llmservice/pricing-and-usage.md — current diff addition -->
| GPT-5.5           |                  5.00 |                  30.00 |                        - |

<!-- gpt-5-5.md — pricing table shows the caveat, but only on the model detail page -->
| **GPT-5.5** | `5.00`  | `30.00`  | 2x input / 1.5x output beyond 272K tokens |

Recommendation

Add a Notes column (or a * footnote) to the global pricing table for GPT-5.5 to surface the long-context surcharge. For example:

| Model             | Input (Credits/Token) | Output (Credits/Token) | Web Search (Credits/Use) | Notes                                      |
| :---------------- | --------------------: | ---------------------: | -----------------------: | :----------------------------------------- |
| ...               |                       |                        |                          |                                            |
| GPT-5.5           |                  5.00 |                  30.00 |                        - | 2× input / 1.5× output beyond 272K tokens |

Apply the same fix to i18n/zh-Hans/.../pricing-and-usage.md.

Minor

[MN-01] GPT-5.5 Document Missing H1 Title (Both Locales)

Description

Every other model documentation file in this repository begins with an H1 heading (# ModelName). The new gpt-5-5.md (both English and Chinese variants) starts directly with ## Overview / ## 概述, skipping the H1. In Docusaurus, the H1 is typically used as the document's page title in the browser tab and as the breadcrumb label. Without it, the page will either inherit the sidebar label as title or render with no H1 at all, breaking visual consistency with peer pages.

Code

<!-- docs/llmservice/models/gpt-5-5.md — line 1 -->
## Overview        ← should be preceded by # GPT-5.5

<!-- All other model pages for reference -->
# DeepSeek V4 Pro   ← deepseek-v4-pro.md line 1
# DeepSeek V4 Flash ← deepseek-v4-flash.md line 1

Recommendation

Add the H1 title as the first line of both files:

# GPT-5.5

## Overview
...

<!-- zh-Hans version -->
# GPT-5.5

## 概述
...

[MN-02] GPT-5.5 Overview States "1-Million-Token Context" but Capabilities Table Shows 1,050,000 Tokens

Description

The Overview paragraph describes GPT-5.5 as having "a 1-million-token context window," while the Capabilities table on the same page shows the actual value as 1,050,000 tokens. The same discrepancy appears in the Chinese translation. This is a minor factual inconsistency within the document itself — the rounded figure in prose is marketing shorthand, but readers who cross-reference these two sections may find the numbers confusing. The gpt-5-5.md Known Limitations section correctly names 1,050,000 tokens as a key detail (the pricing break at 272K).

Code

<!-- Line 3 — Overview -->
GPT-5.5 features a natively omnimodal architecture, a 1-million-token context window...

<!-- Line 26 — Capabilities table -->
| **Context Window** | 1,050,000 tokens (2x pricing for input beyond 272K tokens) |

Recommendation

Align the overview text with the precise figure:

GPT-5.5 features a natively omnimodal architecture, a 1,050,000-token context window...

Or keep the shorthand and add a parenthetical:

...a ~1-million-token context window (1,050,000 tokens)...

Apply the same fix to the Chinese translation.

[MN-03] Uninformative Commit Message

Description

The second commit is titled simply update. This is a one-word message that provides no information about what was changed. Even for a documentation project, commit history is used for tracing changes, changelogs, and automated release notes. The companion commit c198cb6 Sync new model docs from BAI-docs is well-written; 4a1c769 should follow the same standard.

Recommendation

Amend the commit message (or ensure future commits follow a descriptive convention):

docs: add GPT-5.5, DeepSeek V4 Pro/Flash to pricing table and sidebars

Suggestions

[S-01] Clarify Flash-Max Mode Billing Status in DeepSeek V4 Flash Docs

File: docs/llmservice/models/deepseek-v4-flash.md · i18n/zh-Hans/.../models/deepseek-v4-flash.md

Description: The document mentions "V4-Flash-Max" as a reasoning mode activated with a larger thinking budget, and notes it may "increase latency and cost for complex tasks." It's unclear whether Flash-Max is a separate model ID with its own pricing on the platform, or simply a prompt configuration. Users may try to select it as a model and be confused when it doesn't appear.

Suggestion: Add a brief clarification, e.g., "Flash-Max is not a separate model tier on this platform — it refers to providing a larger thinking budget via prompt configuration within the same V4 Flash model."

[S-02] Add Knowledge Cutoff Field to New Model Docs for Consistency

File: docs/llmservice/models/gpt-5-5.md, deepseek-v4-pro.md, deepseek-v4-flash.md (and zh-Hans equivalents)

Description: The existing gpt-5-4.md capabilities table includes a Knowledge Cutoff row (August 31, 2025). None of the three new model pages include this field, which is practically useful for users doing knowledge-intensive tasks.

Suggestion: Add a Knowledge Cutoff row to the Capabilities table in each new model document, or explicitly state "Not disclosed" if unknown.

[S-03] Add Official Source References to DeepSeek V4 Model Pages

File: docs/llmservice/models/deepseek-v4-pro.md, docs/llmservice/models/deepseek-v4-flash.md

Description: The gpt-5-5.md page includes a ## References section with links to the official OpenAI announcement and pricing page. The two new DeepSeek pages contain no references. Authoritative source links help users verify benchmarks and technical claims independently.

Suggestion: Add a references section:

## References

[1] [DeepSeek V4 Announcement](https://deepseek.com/blog/deepseek-v4)
[2] [DeepSeek API Pricing](https://platform.deepseek.com/pricing)

[S-04] Pricing Table — Consider Adding a "Notes" Column for Tier/Limit Context

File: docs/llmservice/pricing-and-usage.md · i18n/zh-Hans/.../pricing-and-usage.md

Description: The current pricing table has only three columns: Input, Output, Web Search. As more models with non-linear pricing (long-context multipliers, minimum spend thresholds) are added, important billing details get buried in per-model pages. This is already an issue for GPT-5.5 (addressed in MJ-01) and could affect future additions.

Suggestion: Add an optional Notes column to the pricing table and populate it with the most important billing caveat for each model (e.g., long-context multipliers, free-tier limits). Models with no special conditions can be marked -.

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 main and ai-bankofai-patch-1. Factual accuracy of third-party benchmark scores and model specifications is taken as provided by the author; independent verification against primary sources is recommended for benchmark claims.

Report generated by Code Review Skill v1.0.0
Date: 2026-04-28