The official SDK turns the launch into an integration

CONFIRMED: OpenAI has turned GPT Image 2.5 from a product announcement into an official integration surface for developers. Version 3.10.0 of its Python SDK, dated September 8, 2026, added the GPT Image 2.5 models and their new image options to the official library. Python teams can now use identifiers and parameters recognized by the client maintained by OpenAI itself, rather than relying on manual requests or waiting for third-party libraries to catch up.

Flare and Sunburst are different bets

This does not announce a third model. It formalizes in the SDK the two variants from ChatGPT Images 2.5: gpt-image-2.5-flare, aimed at fast everyday generation, and gpt-image-2.5-sunburst, designed for workflows where editing precision matters more than latency. A campaign with many adaptations benefits from speed; a product photo or delicate brand edit needs control.

The real gain is the integration contract

The official library is generated from the OpenAPI specification and includes types for request parameters and responses. With the models and options now included, autocomplete, validation, and code maintenance can track the API surface more quickly. That reduces typing mistakes and local adaptations, although teams still need to test limits, output formats, availability, and cost.

Credential expiration arrives in the same package

The same version added expiration fields for service account API keys. This governance detail matters to automated operations because it lets the client represent when a credential is expected to stop being accepted. The release note does not promise automatic rotation. Companies running agents, jobs, or pipelines still need alerts, secure replacement, and a fallback plan. Exposing an expiration date is not the same as managing a key's lifecycle.

What changes for Brazilian businesses

For small and midsize businesses, the gain is shortening the path from testing to putting a visual feature into operation. An agency can build an ad variation generator and switch models according to urgency. An e-commerce team can test catalog edits without maintaining its own type layer. An operation using service accounts gets another signal to avoid discovering an expired credential only when a job fails.

It is not yet proof of production quality

The SDK does not prove consistent quality with Portuguese copy, logos, packaging, or Brazilian references, and the claim of up to 50% lower latency remains an OpenAI statement rather than an independent benchmark. External confirmation of the identifiers shows that the ecosystem recognizes them, but it does not replace a controlled test. Installing version 3.10.0 provides typed access to the surface, not guaranteed commercial access, pricing, or performance for every account.

Verdict

This is a small version update with a meaningful integration signal. Teams already using the image API in Python should test it in isolation, measure latency, cost, and rework, and promote the package only afterward. Teams still prototyping can use the official SDK as a cleaner starting point. The launch becomes useful when it stops being a demo and survives everyday code, credentials, and catalog operations.

Sources

OpenAI, official Python SDK v3.10.0 release: https://github.com/openai/openai-python/releases/tag/v3.10.0 | PyPI, package metadata for openai v3.10.0: https://pypi.org/project/openai/3.10.0/ | Simon Willison, independent confirmation of the IDs and Flare/Sunburst split: https://simonwillison.net/2026/Sep/8/introducing-chatgpt-images-25/ | Traceary, independent SDK history catalog: https://traceary.com/openai-python