go bananas

Appearance

OpenAI GPT Image 2.6 ​

Go Bananas supports both API models OpenAI released in November 2026:

Go Bananas model IDOpenAI API modelBest suited to
openai-gpt-image-2.6-flaregpt-image-2.6-flareFast generation and edits
openai-gpt-image-2.6-sunburstgpt-image-2.6-sunburstDetailed generation and precise edits

OpenAI suggests moving GPT Image 2 and GPT Image 2.5 Flare work to 2.6 Flare, and GPT Image 2.5 Sunburst work to 2.6 Sunburst. GPT Image 2 and both 2.5 models stay available.

Both are pro-tier models. New workspaces can use them straight away, and migration 0062 adds both to every existing workspace's allowed models.

Where 2.6 Flare is the default ​

2.6 Flare is Go Bananas' OpenAI default:

  • New workspaces: a workspace set up with an OpenAI key starts on it, and the key is checked against it. A key that can't use 2.6 is refused at setup instead of on the first image.
  • Keys added later: a key added or replaced in settings is checked against the workspace's own model when that is an OpenAI model, otherwise against 2.6 Flare.
  • Posters: generate_poster without model_id uses 2.6 Flare when the workspace allows it and has an active OpenAI key, then GPT Image 2 on the same terms, then the workspace default. Existing OpenAI workspaces' posters move to 2.6 Flare too; if their key can't use it, the poster fails with OpenAI's reason, and passing model_id: "openai-gpt-image-2" keeps the old model.
  • Composer fallback: the automatic fallback tries 2.6 Flare first among the OpenAI models. 2.6 Sunburst is used only when chosen.

These stay as they were:

  • An existing workspace keeps its default model and its saved choices.
  • A request with a character and no model_id, a character poster included, uses 2.6 Flare when the workspace allows it, has an active OpenAI key, and the request fits it: no Google Search, thinking or Gemini resolution tier, and a ratio, reference count and prompt length 2.6 Flare takes. This is the face default since 9 November 2026. In a face test at full-body framing on 8 November 2026, 2.6 Flare held four of the five approved faces and drew the fifth with a rounder face; GPT Image 2 held all five. To keep a character on GPT Image 2, pass model_id: "openai-gpt-image-2" or save it as the character's preferred model.

Use ​

The workspace's OpenAI key pays for these models, under the usual OpenAI account access and billing. The composer offers both models. The API and MCP tools accept their IDs for generation, editing, references and character preferences.

json
{
  "prompt": "A clean banana logo on a transparent background",
  "model_id": "openai-gpt-image-2.6-flare",
  "quality": "xhigh",
  "complexity": "medium",
  "background": "transparent",
  "output_format": "png",
  "size": "1024x1024"
}

On the CLI, name the model on the provider lane:

sh
gobananas generate "A clean banana logo" --provider --model openai-gpt-image-2.6-flare

Controls ​

The controls are those of GPT Image 2.5, plus complexity:

  • Quality: auto, low, medium, high, xhigh or max.
  • Output: PNG, JPEG or WebP. A transparent background needs PNG or WebP. JPEG and WebP take a compression level from 0 to 100.
  • Size: auto, or custom dimensions that meet all of these:
    • both edges are multiples of 16
    • neither edge is over 3840 pixels
    • the aspect ratio is between 1:3 and 3:1
    • the total is 655,360 to 8,294,400 pixels
  • Complexity: low, medium or high: how much the model plans the picture before rendering it. It is separate from quality and costs nothing extra. Left out, both models plan at high. OpenAI suggests lower complexity for focused edits, such as changing a background or a sign, and higher for dense text, complex layouts and relationships between objects. Higher complexity can take longer and make changes you did not ask for, so compare results. Other OpenAI models refuse it with a 400 before anything is sent to OpenAI; Gemini models ignore it. In the Composer it appears under Advanced Options when a 2.6 model is chosen.
  • Masked edits: not available yet for 2.6 Flare and Sunburst, because OpenAI does not list inpainting for these models and Go Bananas has not checked regional-edit quality. A request with mask_image_id is refused before anything is sent to OpenAI. Use GPT Image 2 for masked edits.

These features of OpenAI's 2.6 models are not available in Go Bananas yet:

  • SVG output
  • animated GIF output
  • markup-guided edits

The provider timeout for these models is 240 seconds.

Cost ​

OpenAI prices 2.6 at the same token rates as 2.5. Tokens per image still differ by model, quality and size. So check_quota gives no dollar figure for these models. It shows a warning, and OpenAI's bill is the true cost.

Official references ​

Checked 7 November 2026: