Skip to content

feat: add gpt-6-1-terra, gemini-3-pro-image, and gemini-3-1-flash-image to AI Gateway /models - #6904

Open
danieltprice wants to merge 1 commit into
mainfrom
docs/ai-gateway-new-model-pages
Open

danieltprice wants to merge 1 commit into
mainfrom
docs/ai-gateway-new-model-pages

Conversation

@danieltprice

Copy link
Copy Markdown
Collaborator

Adds three AI Gateway models to the /models page and their docs pages, so the pages show code examples instead of the "not available" placeholder.

  • capabilities.json: measured entries for gpt-6-1-terra, gemini-3-pro-image, and gemini-3-1-flash-image (from probe:conformance, new rows only)
  • data.json: regenerated with update-models-from-ai-gateway.mjs; adds the three models and replaces $0 prices for kimi-k3-1-pro and gemini-4-carbon with the gateway's rates
  • gemini.md: adds the two image models to the supported models table
  • openai-responses.md: adds gpt-6-1-terra; notes that the Gemini image models generate images on chat completions and the Gemini endpoint
  • models.md: adds the image models to the content shape admonition, with the image_url block shape

NEON PREVIEW

This pull request and its description were written by Isaac.

…ge to AI Gateway /models

Co-authored-by: Isaac <no-reply@databricks.com>
@vercel

vercel Bot commented Nov 20, 2026 •

Copy link
Copy Markdown

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

Project Deployment Actions Updated
neon-next Ready Ready Preview Nov 20, 2026 4:54pm UTC

Request Review

@danieltprice

Copy link
Copy Markdown
Collaborator Author

Live test results

Tested 2026-11-20 against a production AI Gateway branch (aws-us-east-2, Scale plan).

Model /v1/models Chat completions Native endpoint Probe result
gpt-6-1-terra Listed 200, string content Responses 200 conforms; web search and image generation tools work. Function tools on chat completions need reasoning_effort: "none"
gemini-3-pro-image Listed 200, returns an image_url PNG (data URL) Gemini 200 array-content
gemini-3-1-flash-image Listed 200, returns an image_url PNG (data URL) Gemini 200 array-content
  • A non-streaming gemini-3-pro-image response of 2.7 MB returned 200 on chat completions.
  • Checks: check-models-endpoint-sync.js (68 models in sync), model vitest suites (101 passed), verify:agent-endpoints (13/13), generate:ai-gateway-model-markdown (all three pages render code examples).

This pull request and its description were written by Isaac.

@philip philip left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Reviewed by Philip with his agent (live-tested):

What's here is accurate, but I suggest holding the merge until the pricing is complete and the issues below are addressed. Tested on production:

  • /v1/models lists all three, and the data.json text prices and limits match it, including the kimi-k3-1-pro and gemini-4-carbon updates.
  • gpt-6-1-terra: chat returns a string, and Responses, web search, and the page's image examples (AI SDK, TypeScript, Python) work.
  • Both Gemini image models return image_url PNGs on chat completions and inlineData PNGs on the Gemini endpoint.
  • Model tests (149) pass.

Before merging:

  • Pricing: the gateway also bills image output ($120/M for gemini-3-pro-image, $60/M for gemini-3-1-flash-image). The pages show only the text output rate ($12 and $3), so image generation looks 10 to 20 times cheaper than it is. And confirm the overall pricing with PM.
  • Flash Image errors: non-streaming chat requests to gemini-3-1-flash-image returned 200 with finish_reason: "error" and empty content in 4 of 25 tries. Streamed and Gemini endpoint requests all returned images (22 of 22). This needs a fix upstream or a note to stream.
  • Image filter: with imageGeneration: false, the Gemini image models don't appear under the Image filter, and their pages show only text examples.

This branch was successfully deployed

1 active deployment
Preview – neon-next — 79f8cc9d Deployed Nov 20, 2026 by vercel[bot]
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants