Gemini 3 Pro Image
Image Available Full comparison ↗ google/gemini-3-pro-image · by Google DeepMind
· Diffusion transformer (DiT)
Pricing — 1 offering(s)
About this offering: Marketed as Nano Banana Pro: the higher-end model in Google's Gemini-native image generation and editing family, built on Gemini 3 Pro. It's separate from Google's Imagen text-to-image line. The stable API model id is gemini-3-pro-image (some older snippets still use the launch-era -preview alias), and the prices shown are Standard (interactive) rates.
Standard · 1K/2K output
- $0.134 / image
Current
2026-07-19 → present
Billed as image output tokens at $120 per 1M; Google's per-image equivalent is $0.134 for both 1K and 2K…
Billed as image output tokens at $120 per 1M; Google's per-image equivalent is $0.134 for both 1K and 2K output. Batch and Flex cost half: $0.067 per image (Gemini API pricing page, checked 2026-10-11).
Standard · 4K output
- $0.24 / image
Current
2026-07-19 → present
Standard (interactive) rate for 4K output. Batch and Flex cost half: $0.12 per 4K image (Gemini API pricing…
Standard (interactive) rate for 4K output. Batch and Flex cost half: $0.12 per 4K image (Gemini API pricing page, checked 2026-10-11).
Showing the active price and any recorded history. Full pricing history is available via the paid API — see API docs.
See how Gemini 3 Pro Image fits into a cost-aware routing setup
See how →Capability profile
How Gemini 3 Pro Image rates across core capability dimensions, with the task-level evidence behind each rating.
Google DeepMind's own model page claims 'the lowest error rates' in single-line text rendering across languages vs competing models — a vendor comparative claim, not independently benchmarked here.
Built on Gemini 3 Pro's reasoning for visualizing information — DeepMind's own framing (creating accurate infographics, diagrams, data visualization from instructions).
1K/2K/4K output resolution, per DeepMind's own model page.
Supports major edit operations (lighting, color grading, compositional adjustments) per DeepMind's own page, but the same page also lists 'complex image blending operations' as a known struggle — a genuine, vendor-disclosed limitation, not just strength.
Positioned as the higher-quality, slower sibling of Flash Image, but no specific latency claim found to rate against.
Ratings are estimated — limited independent data is available for this model.
Operator guidance
Route here when in-image text accuracy or infographic and diagram generation matters more than speed or cost; Google positions it as its most capable Gemini-native image model. On overall preference, Artificial Analysis's text-to-image arena ranks Google's cheaper Nano Banana 2.1 higher (#4, Elo 1160, vs this model's #13, Elo 1102, checked 2026-10-11), so try Nano Banana 2.1 first when general image quality, speed or cost is the goal.
Use cases
- Text-heavy visuals — posters, diagrams, infographics with multi-language legible text
- Character-consistent series work — up to 5 characters and 14 objects held consistent across generations, per DeepMind's own model page
- Higher-stakes generation/editing where accuracy matters more than speed (the Flash Image sibling is the faster/cheaper choice)
Limitations
- Google DeepMind's own model page says it can struggle with small faces, spelling accuracy and complex image blending
- No independently published GenEval / T2I-CompBench score found — capability ratings above are vendor-stated except where a disclosed weakness is noted
- Google hasn't confirmed the architecture; the diffusion-transformer label is based on secondary technical reporting