Repository navigation
docs(console): document image and voice models - #64473
Conversation
There was a problem hiding this comment.
I checked the new section and tables against the code but didn't run the docs build. Most of it lines up: the /tools#image-generate anchor, the experimental.image_generate.model key, the gpt-live-1 WebRTC /live/sessions route, and the model IDs in the catalog.
One claim doesn't hold up: the suggested opencode/gpt-image-2.5-sunburst value for experimental.image_generate.model (see inline comment).
Public API: unchanged. This only changes docs.
| `gpt-live-1` runs voice sessions. They aren't chat models, so they never appear in `/models` or the models endpoint, | ||
| and `opencode/<model-id>` doesn't select them. Call them on their native API with a Console key, or let OpenCode's | ||
| [`experimental_image_generate`](/tools#image-generate) tool use an image model by setting | ||
| `experimental.image_generate.model`, for example `opencode/gpt-image-2.5-sunburst`. |
There was a problem hiding this comment.
I don't think opencode/gpt-image-2.5-sunburst works here today, and it contradicts the sentence just before it ("opencode/<model-id> doesn't select them").
experimental_image_generate only picks from models.available(["image"]). In the bundled models.dev catalog, the opencode provider has no image models: it has 130 chat and 2 decision models. gpt-image-2.5-sunburst only shows up under openai, azure and aihubmix.
As a configured default, it would be quietly replaced by an automatic pick. Passed explicitly, the tool fails with "is not an available image model".
Unless the live catalog already lists these under opencode, could you drop this pointer or show the native provider form (e.g. openai/gpt-image-2.5-sunburst) instead?
Adds GPT Image 2.5 Sunburst, Nano Banana 2.1, Grok Imagine Image 2.0 and GPT-Live 1 to the Console endpoints and pricing tables, plus a short section on how they're called and used.
Pricing covers per-image, per-token and per-minute units, including Nano Banana 2.1's image-token rates.
Verified with the docs build (
bun run generate && bun src/docs/script/build.ts --out <tmp>).