GPT-Image-2.5 Flare: Pricing, Quality Settings and Speed

What GPT-Image-2.5 Flare is

GPT-Image-2.5 Flare is an image generation and editing model from OpenAI, released on 8 September 2026; its dated snapshot is gpt-image-2.5-flare-2026-09-08. OpenAI's model page marks it as a default model and describes it as its fastest model for high-quality, everyday image generation, rating its performance higher and its speed very fast. It accepts text and image inputs and produces images. On LMArena's image leaderboard GPT-Image-2.5 Flare has 1398 Elo (rank 3).

Quality settings and where to call it

GPT-Image-2.5 Flare supports six quality settings: low, medium, high, xhigh, max and auto, the same set as GPT-Image-2.5 Sunburst. A higher setting spends more output tokens on each image, so it costs more per picture, while auto leaves the choice to the API. The model can be called directly in the Image API, on the generations endpoint for new images and the edits endpoint for changing an existing image, or chosen as the model behind the Responses API image generation tool, so a chat model can produce images during a conversation. OpenAI's image generation guide uses the 2.5 models in its existing code examples, while the features added with the 2.6 models are described in their own sections of the same guide.

Pricing

OpenAI bills GPT-Image-2.5 Flare by tokens rather than per image. Text input costs $5 per million tokens, with cached text input at $1.25. Image input costs $8 per million tokens, with cached image input at $2. Image output costs $30 per million tokens. The number of tokens an image uses depends on its size and quality setting, so the price of a single picture varies with those choices; OpenAI's image generation guide lists the token counts.

Rate limits and safety identifiers

OpenAI does not publish fixed rate limits for GPT-Image-2.5 Flare on its model page. Limits follow the account usage tier, which rises automatically as an organisation sends more requests and spends more on the API, and each account can see its own limits in its settings. For apps that serve individual end users, OpenAI's image generation guide recommends sending a stable, privacy-preserving safety_identifier with each generation or edit request: a hash of a username or email address rather than the address itself, up to 128 characters. The older user parameter on the Images API is deprecated but still accepted, and a non-empty safety_identifier takes precedence over it.

Flare, Sunburst and the 2.6 models

GPT-Image-2.5 Sunburst came out the same day at the same token prices. OpenAI describes it as its most capable image model and rates it highest on performance but medium on speed, where Flare is rated higher and very fast. On 4 November 2026 OpenAI released GPT Image 2.6 as a new Sunburst and Flare at the same prices, adding a complexity field, markup-guided edits, SVG output and animated GIF output. On LMArena GPT-Image-2.5 Sunburst has 1429 Elo (rank 1) and GPT-Image-2.6 Flare has 1384 Elo (rank 4, medium effort). These are LMArena's figures; the site does not recommend one.

Official Resources