GPT-Image-2.5 vs Nano Banana 2 and MAI-Image-2.6

OpenAI is cheapest of the three vendors at $30/M, MAI-Image-2.6 sits at $38/M, and Nano Banana 2 charges $60/M double OpenAI's rate.

All rates are per 1M tokens. Note that input is priced by modality: for the OpenAI models, text input is $5.00/M and image input is $8.00/M, cached input have the same split ($1.25 text, $2.00 image).

What each model is good for

• GPT Image 2.5 Sunburst / Flare / GPT Image 2 - identical pricing across
  every line, so choosing between them is a speed-versus-quality decision, not
  a cost one. Flare is the fast default (roughly half the latency of GPT Image 2
  at comparable quality); Sunburst trades generation time for precision across
  multi-step edits.
• MAI-Image-2.6 - matches OpenAI's $5/M text input and charges nothing for
  non-image output, but the most expensive image output of the two at $38/M.
  No cached-input discount.
• Nano Banana 2 (Gemini 3.1 Flash Image) - by far the cheapest input at
  $0.50/M, which pays off in reference-heavy editing where you upload many
  source images. You give it back at generation time.

Caveats: only the OpenAI models offer cached input, a 75% discount on
repeated prompts or system context. Token counts per image vary with resolution
and quality tier, so none of these rates translate directly into a per-image
cost. It's good to have your own benchmark before committing.

ChatGPT images 2.5 model card