GPT Image 2.5 in Codex: what OpenAI says, and API options
OpenAI says ChatGPT Images 2.5 is available to Codex users. To call GPT Image 2.5 from your own code or agent, Sume offers hosted MCP and a REST API.
Yes, per OpenAI: on September 8, 2026 it said Images 2.5 is available to all ChatGPT, ChatGPT Work and Codex users across desktop, mobile and web. That covers using image generation inside Codex itself. If you instead want to call GPT Image 2.5 from your own script or from an agent you configure, that is the API models, a separate thing.
OpenAI facts come from its launch page and changelog, both read 2026-09-29. Sume facts come from the MCP tools docs and the Image API docs.
What does the Codex changelog say about images?
The changelog's Codex CLI 0.158.0 entry, dated 2026-09-28, lists that image generation and editing can explicitly request transparent backgrounds, and that edits now accept file-backed conversation images. The page does not tie that entry to a specific model, so do not read it as a statement about GPT Image 2.5.
Is the API model the same thing?
OpenAI's page introduces two API models, GPT-Image-2.5 Flare and GPT-Image-2.5 Sunburst, alongside the ChatGPT feature. Flare is described as the default choice for most applications; Sunburst as built for premium workflows that need tighter control across edits. Sume lists them as openai/gpt-image-2.5 (Flare) and openai/gpt-image-2.5-sunburst.
How can I use GPT Image 2.5 from an agent or script?
Two routes on Sume. From code, POST /v1/images with the model id. From an MCP client such as Codex, connect the hosted MCP server at https://mcp.sume.com/mcp; the docs name Cursor, Claude Code and Codex as clients for it. Its generate_image tool is a paid tool.
| Item | What the docs say |
|---|---|
| Endpoint | https://mcp.sume.com/mcp |
| Image tool | generate_image (paid) |
| Required on paid tools | idempotency_key |
| Cost preview | dry_run=true, admission and cost only, no job |
| Model choice | Omit payload.model to route to sume/auto unless the user named a family |
What does the REST call look like?
If you would rather skip MCP, the REST call is one request. It returns 200 with the image URL, or 202 with a job if the render passes 30 seconds. See GPT Image 2.5 request returned 202 for that case.
curl -X POST "https://api.sume.com/v1/images" \
-H "Authorization: Bearer $SUME_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "openai/gpt-image-2.5", "prompt": "a red panda astronaut, studio lighting"}'What should I know before pointing Codex at it?
Codex is one MCP client among several, so the same rules apply as in any other. Four points from the docs, plus one link to keep handy.
- Read-only by default: under OAuth
mcp:read, mutating and paid tools stay hidden until the session hasmcp:writeor an API key. - Spend goes through the wallet and admission; the docs say there is no
mcp:paidscope. - Slow images can outlast a client's tool timeout; see Codex MCP tool timeout.
- To pin GPT Image 2.5 rather than Auto, see GPT Image MCP server.
Sources
Related posts
More in Agents
- MCP tool search: how a long Sume tool list loads in Claude Code
Claude Code loads MCP tools on demand with tool search, which is on by default. What that means for Sume's long hosted tool list and how to prompt for it.
- Runway Brand Kits in MCP, and how Sume handles references
Runway MCP agents can now reference Brand Kits, edited only in Runway's app. Sume's docs list no saved kit; you pass references per request.
- Run the Sume video agent from your backend with Agent Completions
POST /v1/agent/completions runs the same agent as the Sume Agents chat, with tools and media generation, and returns an async run receipt you poll or webhook.
- Safe automation for AI agents that call paid APIs
Keep agents read-only by default, keep secrets out of logs, and on hosted MCP send an idempotency_key, preview with dry_run, and cap with max_spend_usd.
Written by Sume