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 5, 2026 11:35am

[github-actions]

Codex (@codex) review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: c9e886d68e

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]

Update Claude Sonnet 5 cache-write duration prices

When claude-sonnet-5 cache creation is priced with explicit 5m/1h cache-control durations, this update now uses the new base write price of 3.75 but leaves the duration-specific prices at the old 2.5/4 values. Those fields are part of the model schema (input_cache_write_5m_cost_per_mil_tokens and input_cache_write_1h_cost_per_mil_tokens), and the adjacent Vertex entry for the same model uses 3.75/6, so requests through the alias will under-report or under-bill cache writes for those durations.

Useful? React with 👍 / 👎.

[github-actions]

Codex Review Response (PR #910)

Addressed

Update Claude Sonnet 5 cache-write duration prices (P2)

Fixed stale claude-sonnet-5 (anthropic provider) cache write duration prices.
The base input_cache_write_cost_per_mil_tokens was already updated to 3.75, but
the duration-specific fields still had old values:

Field	Before	After
input_cache_write_5m_cost_per_mil_tokens	2.5	3.75
input_cache_write_1h_cost_per_mil_tokens	4	6

These now match the Vertex entry (publishers/anthropic/models/claude-sonnet-5)
and are consistent with every other Claude model in the catalog, where the 5m price
equals the base write price and the 1h price is 1.6x the base.

[github-actions]

Codex (@codex) review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 2b78718524

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 requests that omit max_tokens, openaiParamsToAnthropicMesssageParams uses modelSpec?.max_output_tokens as the value sent to Anthropic, so this catalog value becomes the hard output cap. Anthropic's migration guide documents 128k max output tokens for Claude Sonnet 4.6, so lowering this to 64k will cap long generations that previously relied on the default: https://platform.claude.com/docs/en/about-claude/models/migration-guide

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Keep GLM 5.2 cached-input price at $0.14

For Fireworks GLM 5.2 usage with prompt caching, this catalog value is the per-1M cached-input price used by downstream cost calculations. Fireworks' model page for accounts/fireworks/models/glm-5p2 lists $1.40 / $0.14 / $4.40 per 1M tokens for input/cached input/output, so changing cached input to $0.26 will overstate cached-token costs by about 86%: https://fireworks.ai/models/fireworks/glm-5p2

Useful? React with 👍 / 👎.

[github-actions]

Codex Review Response — PR #910

Issues Addressed

1. Claude Sonnet 4.6 max_output_tokens (P2)

Fixed. Restored max_output_tokens from 64000 to 128000 for:

• claude-sonnet-4-6 (Anthropic provider)
• databricks-claude-sonnet-4-6 (Databricks provider)

Verified against Anthropic's official model documentation at platform.claude.com, which lists 128k max output for Claude Sonnet 4.6. The Bedrock and Vertex entries (anthropic.claude-sonnet-4-6, global.anthropic.claude-sonnet-4-6, us.anthropic.claude-sonnet-4-6, publishers/anthropic/models/claude-sonnet-4-6) already had the correct value of 128000.

2. GLM 5.2 cached-input price (P2)

Fixed. Corrected input_cache_read_cost_per_mil_tokens from 0.26 to 0.14 for accounts/fireworks/models/glm-5p2.

Verified against the Fireworks model page (fireworks.ai/models/fireworks/glm-5p2), which lists pricing as $1.40 / $0.14 / $4.40 per 1M tokens for input / cached input / output. The prior value of $0.26 overstated cached-token costs by ~86%.

Note: The zai-org/GLM-5.2 entry (baseten/together providers) also has 0.26 but was not changed, as those providers may have different pricing and the Codex review only flagged the Fireworks entry.

Files Modified

• packages/proxy/schema/model_list.json — three values corrected as described above

[github-actions]

Codex (@codex) review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 1a8524f826

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 budget

Anthropic documents Sonnet 4.6 as having the same 128k max output as Sonnet 5 in the Sonnet 5 migration guide. This entry is the canonical Anthropic model, and openaiParamsToAnthropicMesssageParams uses modelSpec?.max_output_tokens as the default when callers omit max_tokens, so lowering it to 64k silently halves the response budget for default claude-sonnet-4-6 requests even though the provider supports 128k.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Preserve the GLM 5.2 cached-input price

The current Fireworks serverless pricing lists GLM 5.2 Standard as $1.40 / $0.14 / $4.40 per 1M input/cached/output tokens, and Fireworks maps Standard to accounts/fireworks/models/glm-5p2 in the GLM 5.2 Fast announcement. Changing this cached-input price to 0.26 applies the older/other-provider rate to every cached-token cost calculation for this Fireworks model, overcharging those requests instead of using the documented $0.14 rate.

Useful? React with 👍 / 👎.

[github-actions]

Codex Review Response

Two issues were raised; both were confirmed regressions introduced by the automated model catalog sync and have been fixed.

1. claude-sonnet-4-6 max_output_tokens restored to 128k (P2)

The sync lowered max_output_tokens from 128000 to 64000. This was incorrect — Sonnet 4.6 supports 128k max output. Reverted to 128000.

2. accounts/fireworks/models/glm-5p2 cached-input price restored to 0.14 (P1)

The sync changed input_cache_read_cost_per_mil_tokens from 0.14 to 0.26 for the Fireworks GLM 5.2 model. The documented Fireworks rate is $0.14/1M cached input tokens. Reverted to 0.14.

Note: the zai-org/GLM-5.2 entry (baseten/together providers) was not changed by this PR and retains its existing 0.26 rate, which may reflect different provider pricing.