fix: add Google models gemini-3.1-flash-image-preview +3 more#503
fix: add Google models gemini-3.1-flash-image-preview +3 more#503Caitlin Pinn (cpinn) merged 3 commits intomainfrom
Conversation
|
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: 98dc320c91
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".
| "format": "google", | ||
| "flavor": "embedding", | ||
| "multimodal": true, |
There was a problem hiding this comment.
Do not advertise gemini-embedding-2-preview as a built-in model yet: with format: "google" and flavor: "embedding", /embeddings requests for this model are routed through the Google fetch path, but that path only implements chat-style handling (fetchGoogleChatCompletions expects messages and calls .find on it). Embedding payloads provide input instead, so requests to this newly added model will fail at runtime rather than producing embeddings.
Useful? React with 👍 / 👎.
github-actions
Bot
force-pushed
the
chore/autofix-issue-464
branch
from
98dc320 to
cf2a0bc
Compare
Proxy model verificationVerification completed successfully. Passed: 2 Tested models
Passed models
Failed models
|
Fix a couple of issues with the new job that will attempt to verify models exist - find the vercel bot comment correctly - remove unnecessary local model test - fix permissions for posting comment Sample verification that has failed because of missing credentials: #512 (comment) Update made a few changes as credentials could not be found. New verification that actually did end up failing the models I expected: #503 (comment)
1 participant
fix: add Google models gemini-3.1-flash-image-preview +3 more
Closes #464
Source issue: #464
Summary
gemini-3.1-flash-image-previewgemini-2.5-flash-imagegemini-embedding-2-previewgemini-2.5-computer-use-preview-10-2025gemini-3.1-flash-image-previewgemini-2.5-flash-imagegemini-embedding-2-previewgemini-2.5-computer-use-preview-10-20252
3
4
5
6
7
8
Verified metadata
Verification notes
Verification
Sources and fields verified
sync_models (LiteLLM) cross-check
All four models (
gemini-3.1-flash-image-preview,gemini-2.5-flash-image,gemini-embedding-2-preview,gemini-2.5-computer-use-preview-10-2025) are entirely absent from the LiteLLMmodel_prices_and_context_window_backup.jsoncatalog. Therefore every proposed field deviates from sync_models (which has no data). All values are sourced from official Google documentation as listed above. No sync_models values exist to compare against.Per-model field notes
gemini-3.1-flash-image-preview
format: "google" — inferred from provider (Google), consistent with all existing Gemini entriesflavor: "chat" — matches existinggemini-3-pro-image-preview(image gen models use chat flavor)multimodal: true — accepts text and image inputs per Vertex docsinput_cost_per_mil_tokens: 0.5 — Google AI pricing page and Vertex pricing page both confirm $0.50/Moutput_cost_per_mil_tokens: 3 — text output pricing; image output is $60/M (separate billing dimension not expressible in schema)max_input_tokens: 131072 — Vertex AI model pagemax_output_tokens: 32768 — Vertex AI model pagesupported_regions: ["global"] — Vertex AI model pageavailable_providers: ["google", "vertex"] — confirmed on both Google AI docs and Vertex AI model pagedisplayName: "Gemini 3.1 Flash Image Preview" — follows style of existinggemini-3-pro-image-preview("Gemini 3 Pro Image Preview")gemini-2.5-flash-image
format: "google"flavor: "chat" — matches existing image generation model patternmultimodal: true — accepts text and image inputs per Vertex docsinput_cost_per_mil_tokens: 0.3 — confirmed on both pricing pagesoutput_cost_per_mil_tokens: 2.5 — text output pricing per Vertex pricing page; image output is $30/M (separate billing dimension)max_input_tokens: 32768 — Vertex AI model pagemax_output_tokens: 32768 — Vertex AI model pagesupported_regions: ["global"] — Vertex AI model page lists global plus regional endpointsavailable_providers: ["google", "vertex"] — Vertex docs confirm GA availabilitydisplayName: "Gemini 2.5 Flash Image"gemini-embedding-2-preview
format: "google"flavor: "embedding" — this is an embedding model; ModelFlavors includes "embedding"multimodal: true — accepts text, image, video, audio, and PDF inputs per dedicated model docsinput_cost_per_mil_tokens: 0.2 — text input pricing from Google AI pricing page; image/audio/video input have different rates ($0.45, $6.50, $12.00/M respectively) but schema only supports a single input rateoutput_cost_per_mil_tokens: omitted — embedding models produce vectors, not tokensmax_input_tokens: 8192 — dedicated model docs pagemax_output_tokens: omitted — not applicable for embedding modelsavailable_providers: ["google", "vertex"] — embeddings docs explicitly state "Available through Gemini API or Vertex AI"displayName: "Gemini Embedding 2 Preview"gemini-embedding-2-previewmatching the dedicated docs page URL.gemini-2.5-computer-use-preview-10-2025
format: "google"flavor: "chat"multimodal: true — accepts image inputs (screenshots) per model docsinput_cost_per_mil_tokens: 1.25 — confirmed on both Google AI and Vertex pricing pages; matches Gemini 2.5 Pro standard-tier pricing. Note: Google also lists a higher rate ($2.50/M) for prompts >200k tokens, but the schema only supports a single rateoutput_cost_per_mil_tokens: 10 — confirmed on both pricing pages; matches Gemini 2.5 Pro. Higher rate ($15/M) for prompts >200k tokens not expressible in schemamax_input_tokens: 128000 — dedicated model docs pagemax_output_tokens: 64000 — dedicated model docs pageavailable_providers: ["google", "vertex"] — listed on Vertex AI pricing page and available in Vertex model gardendisplayName: "Gemini 2.5 Computer Use Preview (10-2025)"Fields not published or not applicable
sync_models vs proposed update
sync_models cross-check found differences. Official provider verification was used for the applied values, and sync_models discrepancies are listed below for review.