chore: remove deprecated models from the catalog - #986
Conversation
6c6051f to
3e94c8c
Compare
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 3e94c8cd2d
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".
| "magistral-small-2506", | ||
| "meta-llama/Meta-Llama-3-8B-Instruct-Turbo", | ||
| "meta-llama/llama-4-maverick-17b-128e-instruct", | ||
| "meta-llama/llama-4-scout-17b-16e-instruct", |
There was a problem hiding this comment.
When the proxy is used with Groq enterprise/committed-spend credentials, this model is still documented as available after the July 17, 2026 free/developer-tier shutdown: Groq's deprecation page says enterprise customers with a committed-spend contract are not affected, and the model card still shows a Quick Start for meta-llama/llama-4-scout-17b-16e-instruct (https://console.groq.com/docs/deprecations, https://console.groq.com/docs/model/meta-llama/llama-4-scout-17b-16e-instruct). Adding it to the sync exclusion list, together with removing it from model_list.json/AvailableEndpointTypes, makes getModelEndpointTypes(...) return no Groq route and prevents future syncs from restoring it, so those enterprise requests are incorrectly blocked instead of merely deprecated for non-enterprise use.
Useful? React with 👍 / 👎.
| "qwen-2.5-32b", | ||
| "qwen-2.5-coder-32b", | ||
| "qwen-qwq-32b", | ||
| "qwen/qwen3-32b", |
There was a problem hiding this comment.
When the proxy is used with Groq enterprise/committed-spend credentials, this ID is not a definitive provider-wide removal: Groq's deprecation page scopes the July 17, 2026 shutdown to free/developer usage and says committed-spend enterprise customers are not affected, while the model card still documents qwen/qwen3-32b usage (https://console.groq.com/docs/deprecations, https://console.groq.com/docs/model/qwen/qwen3-32b). Because this addition keeps sync_models from re-adding the catalog entry and the commit also removes its Groq endpoint mapping, enterprise users who can still access the model lose routing through the proxy.
Useful? React with 👍 / 👎.
The audit removed meta-llama/llama-4-scout-17b-16e-instruct and qwen/qwen3-32b because they left the CI account's serverless Groq model list after Groq's 2026-07-17 free/developer-tier shutdown. But per Groq's own docs the shutdown does NOT affect enterprise/committed-spend customers, and both model cards still document usage (console.groq.com/docs/deprecations + the per-model cards) — so removing them (and blacklisting them from re-sync) blocks routing for enterprise users. Restore both to the catalog + index and drop them from deprecated_model_ids.json. Addresses the two codex P2s. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
3e94c8c to
b981a17
Compare
b981a17 to
58192d7
Compare
58192d7 to
0fa6b44
Compare
0fa6b44 to
447ce31
Compare
…aintrustdata/braintrust-proxy into chore/model-deprecation-audit
Single catalog PR containing every genuine "update model catalog" change from the week (#989/#1000/#1003/#1008/#1014/#1017/#1021/#1024), rebuilt on current main after #1009 (sync) and #986 (deprecation) merged. 28 new models + 22 metadata/deprecation-marking changes, with prior review fixes: gemma-3 converse format, gpt-5.6 bedrock cache-write, databricks-glm-5-2 1M context, vertex llama-4 maas pricing/quota, gpt-oss-20b groq-only pin, proxy.ts OpenAI/xAI vertex-maas OpenAPI routing. Also restores GLM-5.2 cached-input to $0.26 (Baseten/Together) — the merged sync left main at Fireworks' $0.14. Validated: gemma-3-4b/12b + nemotron + GLM-5.2 invoke 200 via gateway. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Catalog: - openai/gpt-oss-20b: restore Groq pricing $0.075/$0.30 and Groq-only routing (sync reverted it to Together's $0.05/$0.20 + re-added together). Groq is the priced/routable provider. - zai-org/GLM-5, moonshotai/Kimi-K2.5: drop the re-added `baseten` provider — Baseten deprecated both (410); #986 removed it. Keep `together` (matches index.ts). Resolves the codex P1 (model_list/index.ts drift). - thinkingmachines/inkling: drop re-added `together` (Together's id is case-sensitive `thinkingmachines/Inkling`; bare-id route fails). - zai-org/GLM-5.2: restore cached-input to $0.26 (Baseten/Together; Fireworks' $0.14 belongs to glm-5p2). Durable pins (SYNC_PRESERVED_FIELDS), so these stop recurring each sync: - openai/gpt-oss-20b -> input/output cost (mirrors the existing gpt-oss-120b pin) - zai-org/GLM-5.2 -> input_cache_read Validated: gpt-oss-20b (groq) + inkling (baseten) invoke 200 on the fixed catalog. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
1 participant
Model deprecation audit
Each model below returned a definitive not-found / deprecated response from its provider (a single 404 / not-found is treated as authoritative). Account/region-scoped providers (bedrock, vertex, databricks) are surfaced for manual review, not auto-removed.
{ "deprecations": [ { "model": "moonshotai/Kimi-K2.5", "provider": "baseten", "reason": "absent from list + probe not-found", "status": 410, "detail": "{\"error\":\"the model version you are trying to access has been deprecated.\"}\n" }, { "model": "nvidia/Nemotron-120B-A12B", "provider": "baseten", "reason": "absent from list + probe not-found", "status": 410, "detail": "{\"error\":\"the model version you are trying to access has been deprecated.\"}\n" }, { "model": "zai-org/GLM-5", "provider": "baseten", "reason": "absent from list + probe not-found", "status": 410, "detail": "{\"error\":\"the model version you are trying to access has been deprecated.\"}\n" }, { "model": "zai-org/GLM-5.1", "provider": "baseten", "reason": "absent from list + probe not-found", "status": 410, "detail": "{\"error\":\"the model version you are trying to access has been deprecated.\"}\n" }, { "model": "meta-llama/llama-4-scout-17b-16e-instruct", "provider": "groq", "reason": "absent from list + probe not-found", "status": 404, "detail": "{\"error\":{\"message\":\"The model `meta-llama/llama-4-scout-17b-16e-instruct` does not exist or you do not have access to it.\",\"type\":\"invalid_request_error\",\"code" }, { "model": "qwen/qwen3-32b", "provider": "groq", "reason": "absent from list + probe not-found", "status": 404, "detail": "{\"error\":{\"message\":\"The model `qwen/qwen3-32b` does not exist or you do not have access to it.\",\"type\":\"invalid_request_error\",\"code\":\"model_not_found\"}}\n" } ], "reportOnly": [ { "provider": "bedrock", "modelCount": 122 }, { "provider": "databricks", "modelCount": 36 }, { "provider": "fireworks", "modelCount": 266 }, { "provider": "vertex", "modelCount": 122 } ], "skipped": [ { "provider": "cohere", "reason": "no adapter" }, { "provider": "lepton", "reason": "no adapter" }, { "provider": "openrouter", "reason": "no adapter" }, { "provider": "replicate", "reason": "no adapter" } ] }