chore: sync new models

[github-actions]

Automated sync of model metadata and newly available models from the approved provider allowlist.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
ai-proxy	Ready	Preview, Comment	Jul 8, 2026 1:21pm

[github-actions]

Codex (@codex) review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: ac58641bc7

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "Codex (@codex) review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "Codex (@codex) address that feedback".

[chatgpt-codex-connector]

Keep Sonnet 4.6 at its 128k output limit

For claude-sonnet-4-6, Anthropic documents 128k max output tokens (https://platform.claude.com/docs/en/about-claude/models/migration-guide), and the proxy uses modelSpec.max_output_tokens as the fallback max_tokens when clients omit one (packages/proxy/src/providers/anthropic.ts). With this lowered to 64k, default requests through the proxy will be capped at half of the supported output length.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Keep Fireworks GLM 5.2 cache reads at $0.14

For accounts/fireworks/models/glm-5p2, Fireworks' own pricing table lists GLM 5.2 Standard as $1.40 / $0.14 / $4.40 per 1M tokens (https://fireworks.ai/blog/glm-5p2-fast), so copying the 0.26 cached-input price here makes catalog consumers overstate cached-input cost by about 86% for the Fireworks route.

Useful? React with 👍 / 👎.

[github-actions]

Codex Review Response

Issues Addressed

1. claude-sonnet-4-6 max_output_tokens (P2 - Fixed)

Codex correctly identified that claude-sonnet-4-6 had max_output_tokens: 64000, but Anthropic documents 128k for this model. Verified against https://platform.claude.com/docs/en/docs/about-claude/models — the legacy models table confirms Claude Sonnet 4.6 has a 128k max output.

The Bedrock and Vertex variants already had the correct value of 128000. Fixed the two entries that were wrong:

• claude-sonnet-4-6 (primary/Anthropic API entry): 64000 -> 128000
• databricks-claude-sonnet-4-6: 64000 -> 128000

2. Fireworks GLM 5.2 cached input price (P2 - Fixed)

Codex correctly identified that accounts/fireworks/models/glm-5p2 had input_cache_read_cost_per_mil_tokens: 0.26, but Fireworks' pricing lists GLM 5.2 Standard cached input at $0.14/1M tokens. Verified against https://fireworks.ai/blog/glm-5p2-fast.

Fixed:

• accounts/fireworks/models/glm-5p2: 0.26 -> 0.14

Note: The zai-org/GLM-5.2 entry (baseten/together providers) was not changed as it was not flagged and those providers may have different pricing.

Files Modified

• packages/proxy/schema/model_list.json — three edits (two max_output_tokens fixes, one cached input price fix)
• packages/proxy/schema/index.ts — no changes needed

[github-actions]

Codex (@codex) review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 025b15c0f2

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "Codex (@codex) review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "Codex (@codex) address that feedback".

[chatgpt-codex-connector]

Restore Sonnet 4.6's 128k output limit

For claude-sonnet-4-6, Anthropic's current migration docs state that the 128k max-output limit is unchanged from Sonnet 4.6 (docs), but this entry now advertises only 64k. The proxy uses modelSpec?.max_output_tokens as the Anthropic default when clients omit max_tokens (packages/proxy/src/providers/anthropic.ts), so Sonnet 4.6 requests that rely on the default can be capped/truncated at 64k even though the provider allows 128k.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Keep GLM 5.2 cache-read pricing at $0.14

Fireworks' serverless pricing table lists GLM 5.2 Standard at $1.40 / $0.14 / $4.40 for input/cached-input/output, while $0.26 is the cached-input price for GLM 5.1 (docs). Updating accounts/fireworks/models/glm-5p2 to 0.26 therefore overstates cached-token costs for this model in any metadata-driven cost display or accounting path.

Useful? React with 👍 / 👎.

[github-actions]

Codex Review Response — PR #927

Issues addressed

1. Claude Sonnet 4.6 max_output_tokens (P2) — Fixed

Codex correctly identified that claude-sonnet-4-6 had max_output_tokens: 64000 instead of 128000. Verified against Anthropic's model documentation, which lists Claude Sonnet 4.6 at 128k max output tokens.

Changes made in model_list.json:

• claude-sonnet-4-6 (line 1658): 64000 → 128000
• databricks-claude-sonnet-4-6 (line 11924): 64000 → 128000

Note: The Bedrock variants (us.anthropic.claude-sonnet-4-6, global.anthropic.claude-sonnet-4-6, anthropic.claude-sonnet-4-6) and Vertex variant (publishers/anthropic/models/claude-sonnet-4-6) already had the correct value of 128000.

2. GLM 5.2 cached-input pricing (P2) — Fixed

Codex correctly identified that GLM 5.2's input_cache_read_cost_per_mil_tokens was set to 0.26 (which is the GLM 5.1 price) instead of the correct 0.14. Verified against Fireworks serverless pricing.

Changes made in model_list.json:

• accounts/fireworks/models/glm-5p2 (line 9201): 0.26 → 0.14
• zai-org/GLM-5.2 (line 9215): 0.26 → 0.14

Files modified

• packages/proxy/schema/model_list.json — 4 field value corrections (no structural changes)