Nano Banana 2.1 vs GPT Image 2.5: Price per Image, Quality and Where to Use Each

At 1K, Nano Banana 2.1 costs more than GPT Image 2.5 medium and less than high

Both vendors charge $30 per 1M image-output tokens, but they turn tokens into images differently. Google fixes the token count by resolution, so each Nano Banana 2.1 size has one price. OpenAI's token count depends on the quality setting and size, and its calculator gives one combined estimate for Sunburst and Flare. The figures below are each vendor's published per-image numbers for Standard processing, image output only.

Read across a row: at 1K and 2K, Nano Banana 2.1 sits between GPT Image 2.5's medium and high settings, and at 4K it costs a little more than high. That last row compares different images, though, because Google's 4K price is for a 4096×4096 output while OpenAI caps the total at 8,294,400 pixels and calls anything above 2560×1440 experimental. On OpenAI, a 1536×1024 image costs less than a square one (high $0.0412 against $0.0527).

For a run of 1,000 images at 1K, that is $33.60 on Nano Banana 2.1 against $13.17 at GPT Image 2.5 medium, $52.68 at high and $210.72 at max (ToolColumn arithmetic, image output only). Batch halves image output on both: 2.1 drops to $0.0168 per 1K image, and GPT Image 2.5 to $15 per 1M output tokens, about $0.026 per high-quality 1024×1024 image by ToolColumn's arithmetic.

The per-image figures leave out different extras. OpenAI's exclude text input ($5 per 1M tokens), image input ($8 per 1M) and streaming, where each partial image adds 100 output tokens. Google bills input at $1.50 per 1M and thinking text at $7.50 per 1M, does not charge for the interim thought images, and charges $14 per 1,000 Google Search grounding requests after 5,000 free a month.

Within Google's lineup, Nano Banana 2 Lite matches 2.1's $0.0336 at 1K but makes only 1K images and is not built for multiple references or multi-turn edits. Nano Banana 2 costs $0.067 at 1K and is now Google's previous generation; it is the only one with a 512px size, at $0.045.

No vendor benchmark puts the two side by side

Google's Nano Banana 2.1 model card reports human side-by-side preference only against Google's own models, and OpenAI publishes no quality benchmark for GPT Image 2.5. OpenAI's numbers are about speed: Flare delivers higher-quality images than GPT Image 2 at 50% lower latency, and ChatGPT Images 2.5 cuts generation latency by up to 50% against Images 2.0.

The two vendors list overlapping weak spots. Google's card names small text, long paragraphs and character consistency among 2.1's limits, plus occasional left/right confusion; OpenAI lists text placement, character consistency and composition control, and warns that complex prompts can take up to 2 minutes. For posters, labels and recurring characters, run your own prompts on both before committing a budget.

Sunburst and Flare: same token rates, no separate per-image price

OpenAI calls Flare the default choice for most applications and aims Sunburst, which its model page calls its most capable image model, at premium workflows that need tighter control across edits. Both are priced like GPT Image 2 ($5 text input, $8 image input and $30 output per 1M tokens), both accept the low, medium, high, xhigh, max and auto quality settings, and both have the same default rate limits. The catch is cost per image: OpenAI says token use can differ by model and quality, yet publishes only the combined estimate in the table above, so neither variant's own per-image cost is published.

GPT Image 2 (gpt-image-2) remains available at the same rates. In ChatGPT, OpenAI names no Sunburst or Flare picker and does not document which variant answers a prompt.

Nano Banana 2.1 takes 14 reference images; GPT Image 2.5 does transparent backgrounds

• References and editing: Nano Banana 2.1 fuses up to 14 reference images, with up to 10 objects and 4 characters, and supports multi-turn editing. GPT Image 2.5 edits through OpenAI's edits endpoint, inpainting included.
• Output control: GPT Image 2.5 has five quality levels plus auto, a transparent-background setting and sizes up to 3840 pixels on an edge. Nano Banana 2.1 outputs 1K (the default), 2K or 4K in aspect ratios from 1:8 to 8:1, and offers minimal, medium and high thinking; seed and temperature return an API error.
• Grounding: Nano Banana 2.1 can ground images in Google Search and Google Image Search; Google offers Image Search grounding only on 2.1 and Nano Banana 2.
• Provenance and ownership: Every Gemini API image carries a SynthID watermark, and the Gemini app adds a visible one. OpenAI uses C2PA metadata plus Google DeepMind's SynthID. OpenAI's terms say you own the output; Google's say it won't claim ownership.

Nano Banana 2.1 vs Nano Banana Pro: Pro still fits style references and brand work

Google recommends 2.1 for all new projects and describes it as the more efficient counterpart to Nano Banana Pro. On the 2.1 model card, 2.1 leads Pro on every row Google reports; the figures are Google's.

Read the gap as Google's own measurement: the card puts 2.1 between 77 and 136 Elo points above Pro on the preference and editing rows. With thinking off, the card puts 2.1 at 1015 on overall preference, still above Pro.

Pro costs $0.134 per 1K or 2K image and $0.24 at 4K, about four times 2.1's price at 1K and twice at 4K (ToolColumn arithmetic). It is still the better pick in two cases. Pro is the only Nano Banana model that takes style reference images (up to 3), and it holds up to 5 characters against 2.1's 4, though 2.1 takes up to 10 objects against Pro's 6. Google also keeps Pro as its premium choice for world knowledge, localization, brand consistency and precise creative control. In the Gemini app, paid plans can redo a 2.1 image with Pro, and in Google Flow Pro is the default for AI Ultra.

Where each model runs: Gemini, ChatGPT and the two APIs

• Gemini app: Nano Banana 2.1 is on Google AI Plus ($4.99 a month), Pro ($19.99) and Ultra ($99.99 or $199.99); Free uses Nano Banana 2. Google's help-center limits table still lists only Nano Banana 2, so its pages disagree. Google publishes no image counts, only multipliers (Plus 2x, Pro 4x) and a limit that refreshes every 5 hours up to a weekly cap.
• ChatGPT: ChatGPT Images 2.5 is on every tier, from Free to Go ($8), Plus ($20) and Pro (from $100), and also in ChatGPT Work and Codex. Images with thinking needs Plus, Pro or Business. OpenAI says image limits did not change with 2.5 and publishes no per-plan counts.
• Other Google products: 2.1 is the standard model in Google Flow, where plans include 200 (Plus), 1,000 (Pro) or 10,000 to 25,000 (Ultra) credits; Google does not publish credits per image. Its model card also lists Search AI Mode, Google Ads and Stitch.
• Nano Banana 2.1 API: gemini-nano-banana-2.1 is stable in the Gemini API and Google AI Studio, paid tier only, and generally available on Gemini Enterprise Agent Platform (formerly Vertex AI) in the global region at the same Standard rates. Agent Platform's pricing page lists a Flex rate yet says 2.1 does not support Flex.
• GPT Image 2.5 API: gpt-image-2.5-flare and gpt-image-2.5-sunburst work through the Image API's generation, edit and batch endpoints and as the model behind the Responses API image tool. Since October 28, 2026, eligible ChatGPT Pro and Business subscribers can activate promotional API credits. The older gpt-image-1 shut down on October 23, 2026.

For per-plan app limits, see ChatGPT image generation limits by plan and Gemini usage limits by plan; for every Nano Banana and GPT Image rate, see Nano Banana and GPT Image.