whatmodel.sh / image / Nano Banana 2 (Gemini 3.1 Flash Image)
Nano Banana 2 (Gemini 3.1 Flash Image)
↗ share
+ contribute
Nano Banana 2 (Gemini 3.1 Flash Image) is a closed-weights image model from Google, released February 2026. Licence: proprietary.
Score within the LMArena text-to-image comparison set only.
$ how it compares
The image models ranked either side of Nano Banana 2 (Gemini 3.1 Flash Image), out of 197 with a score, then more from Google.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 7 |  Reve 2.1 | 78±2 | api $0.25/image | ||
| 8 |  Muse Image | 76±2 | api $0.010/image | ||
| 9 | Nano Banana 2 (Gemini 3.1 Flash Image) | 76±1 | api $0.067/image | ||
| 10 | Nano Banana Pro (Gemini 3 Pro Image) | 76±1 | api $0.13/image | ||
| 11 |  Qwen-Image-3.0-Pro | 75±2 | api $0.040/image | ||
| more from Google | |||||
| 5 | Nano Banana 2.1 | 81±2 | api $0.034/image | ||
| 16 | Nano Banana 2 Lite (Gemini 3.1 Flash-Lite Image) | 73±2 | api $0.034/image | ||
| 39 | Nano Banana (Gemini 2.5 Flash Image) | 62±1 | api $0.039/image | ||
$ what it's good at
text to image▂▅▅▃▄▄█▃▁▃▁▁ #7 of 126 · 37% of its score · 5 tests
Beats Muse Image here, though they rank higher overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| LMArena text-to-image | 1261.8 ✎ | ▂▂▃▄▇█▄▂▄▂·▂ beats 66/75 | 23% |
| GenExam (relaxed overall, GPT-5 judge) | 92.6% ✎ | ▆▆·▂▂····▂▃█ beats 9/13 | 5.6% |
| ImageTime (prompt-only, overall) | 74.5% ✎ | ▄·····▄█▄·▄█ beats 6/7 | 3.5% |
| UltraText Bench (Composite, bilingual dense text rendering) | 97.0% ✎ | ▁▂▂█▃▁▂▃·▃·▁ beats 21/22 | 2.5% |
| Qwen-Image-Bench (Q-Judger, v1) | 59.8% ✎ | █▅▅·▅▅█·█··▃ beats 16/17 | 2.4% |
image to image / editing▁▁·▃▂▃█▃▃▄▁▁ #10 of 62 · 26% of its score · 4 tests
Beats Reve 2.1 here, though they rank higher overall. Trails ChatGPT Image Latest (chatgpt-image-latest), Seedream 5.0 Pro, Nano Banana Pro (Gemini 3 Pro Image) here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Edit-Compass (EN, overall) | 74.8% ✎ | ▃▃▂·▃▆▂█·▃·▃ beats 19/20 | 14% |
| Arena image edit | 1387.4 ✎ | ▂▁▁▂▆▇█▄▆▅▁▂ beats 43/53 | 8.9% |
| MultiTextEdit cross-lingual text-in-image editing (SemAvg, LVM judge) | 80.1% ✎ | ▄·▄▄▄▄··█▄·▄ beats 8/8 | 1.9% |
| TECCI human overall success rate (IRCS+GGIS) lab | 19.7% ✎ | ▄········▄█▄ beats 2/4 | 1.3% |
other tests5 not in a task · 36% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| InfoEdit infographic editing (average success rate) | 38.0% ✎ | ▄··▄█····▄▄▄ beats 4/6 | 20% |
| VVRBench-Challenge accuracy (720 harder verifiable tasks) | 3.9% ✎ | ▄·▄···█·▄▄·▄ beats 2/6 | 6.2% |
| Rapidata Benchmark.ai image benchmark (overall Elo) | 1128.1 ✎ | ▁▁▁▄▂█▂▁▁▃▁▁ beats 29/34 | 3.6% |
| PaintBench precise editing (mIoU, 20-task average) | 17.1% ✎ | ▃·····█▅▃·▃▅ beats 9/9 | 3.6% |
| EdiVal multi-turn editing overall (EdiVal-O after turn 3) | 58.5% ✎ | ▂·▂▃·▂·▃▂·▃█ beats 12/14 | 2.9% |
$ where to run it
Through an API
api · $0.067/image · ai.google.dev ↗ · price sourceGoogle Gemini Developer API Standard paid tier; 1K generated image output only. Input/text/thinking/tool charges additional. Product: Gemini Developer API. API model: gemini-3.1-flash-image. Settings: tier=Standard paid, resolution=1K, batch=no, n=1. Price basis: provider-published approximate image OUTPUT cost. Additional/excluded charges: Input tokens, text/thinking output tokens and optional tools. Currency: USD.
$ how the score was worked out · LMArena leaderboard dataset 32%, InfoEdit: Probing Global Layo… 20%, Edit-Compass & EditReward-Com… 14% and 10 more
Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works
$ improve this page
Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.
data updated · sources and licences · json