Update Neuralwatt models: remove dead GLM 5.1 entries, add GLM 5.2 variants (fast, short, short & fast), and add cache-read pricing

[bakhtiar-id]

Summary

This updates the neuralwatt provider catalog to match the requested narrow refresh scope:

• removes deprecated/dead GLM 5.1-era Neuralwatt entries
• adds the missing GLM 5.2 variants
• adds cache_read pricing to all remaining Neuralwatt model configs
• updates the Neuralwatt README to reflect the current model lineup and cache pricing policy

Existing non-GLM Neuralwatt IDs were intentionally left unchanged.

What Changed

Removed dead Neuralwatt GLM entries

Deleted these model configs:

• providers/neuralwatt/models/glm-5-fast.toml
• providers/neuralwatt/models/glm-5.1-fast.toml
• providers/neuralwatt/models/zai-org/GLM-5.1-FP8.toml

Added new GLM 5.2 variants

Added these model configs:

• providers/neuralwatt/models/glm-5.2-fast.toml
• providers/neuralwatt/models/glm-5.2-short.toml
• providers/neuralwatt/models/glm-5.2-short-fast.toml

Configured pricing and limits from Neuralwatt’s live catalog:

• GLM 5.2 family pricing:
  • input = 1.45
  • output = 4.5
  • cache_read = 0.3625
• Short variants use:
  • context = 199_984
  • output = 199_984

Updated cache-read pricing across remaining Neuralwatt models

Added cache_read to all remaining Neuralwatt model TOMLs using the documented Neuralwatt 25% cache-read rate:

• GLM 5.2 family: 0.3625
• Kimi K2.5 family: 0.13
• Kimi K2.6 family: 0.1725
• Kimi K2.7 Code: 0.2375
• Qwen3.5 397B family: 0.1725
• Qwen3.6 35B family: 0.0725

cache_write was intentionally left unchanged.

Reasoning Options Note

GLM 5.2 reasoning options remain:

• high
• max

This matches Neuralwatt’s chat-completions docs, which describe high and max as the native GLM 5.2 reasoning levels, with lower OpenAI-style values normalized onto them.

Validation

Validated successfully with:

bun validate

Sources

Neuralwatt models API: https://api.neuralwatt.com/v1/models
Neuralwatt models docs: https://portal.neuralwatt.com/docs/api/models
Neuralwatt chat completions docs: https://portal.neuralwatt.com/docs/api/chat-completions