whatmodel.sh / image / GPT Image 2
GPT Image 2
↗ share
+ contribute
GPT Image 2 is a closed-weights image model from OpenAI, released April 2026. Licence: proprietary.
Score within the LMArena text-to-image comparison set only.
$ how it compares
The image models ranked either side of GPT Image 2, out of 197 with a score, then more from OpenAI.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 1 | GPT Image 2.5 Sunburst | 92±2 | api $0.013/image | ||
| 2 | GPT Image 2.5 Flare | 88±2 | api $0.013/image | ||
| 3 | GPT Image 2 | 85±1 | api $0.053/image | ||
| 4 |  MAI-Image-2.6 | 81±2 | comparison only | ||
| 5 |  Nano Banana 2.1 | 81±2 | api $0.034/image | ||
| more from OpenAI | |||||
| 14 | ChatGPT Image Latest (chatgpt-image-latest) | 74±4 | api $0.034/image | ||
| 17 | GPT Image 1.5 | 71±1 | api $0.034/image | ||
| 42 | GPT Image 1 | 61±1 | api $0.042/image | ||
$ what it's good at
text to image▂▅▅▃▄▄█▃▁▂▁▃ #3 of 126 · 32% of its score · 6 tests
| test | result | vs the others in its run | counts |
|---|---|---|---|
| LMArena text-to-image | 1383.0 ✎ | ▂▂▃▄▇█▄▂▄▂·▃ beats 73/75 | 22% |
| GenExam (relaxed overall, GPT-5 judge) | 93.8% ✎ | ▆▆·▂▂····▂▂█ beats 11/13 | 2.6% |
| ImageTime (prompt-only, overall) | 78.6% ✎ | ▄·····▄█▄·▄█ beats 7/7 | 2.5% |
| FEPBench instruction faithfulness | 66.4% ✎ | ▄·█···▄▄▄▄▄▄ beats 8/8 | 2.4% |
| Qwen-Image-Bench (Q-Judger, v1) | 64.7% ✎ | █▅▅·▅▅█·█··▃ beats 17/17 | 2.1% |
| UltraText Bench (Composite, bilingual dense text rendering) | 99.4% ✎ | ▁▂▂█▃▁▂▃·▁·▃ beats 22/22 | 1.0% |
image to image / editing▁▁·▃▂▃█▃▃▄▃▁ #3 of 62 * · 8.3% of its score · 1 test
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Arena image edit | 1461.7 ✎ | ▂▁▁▂▆▇█▄▆▅▃▂ beats 51/53 | 8.3% |
other tests6 not in a task · 59% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| InfoEdit infographic editing (average success rate) | 62.0% ✎ | ▄··▄█····▄▄▄ beats 6/6 | 19% |
| VVRBench accuracy (10,000 verifiable synthetic-scene tasks) | 86.9% ✎ | ▃·█▅▃·▅▃···▃ beats 10/10 | 19% |
| VVRBench-Challenge accuracy (720 harder verifiable tasks) | 10.3% ✎ | ▄·▄···█·▄▄·▄ beats 5/6 | 10% |
| PaintBench precise editing (mIoU, 20-task average) | 16.3% ✎ | ▃·····█▅▃·▃▅ beats 8/9 | 4.2% |
| Rapidata Benchmark.ai image benchmark (overall Elo) | 1167.3 ✎ | ▁▁▁▄▂█▂▁▁▂▃▁ beats 32/34 | 3.4% |
| ChartSync chart editing (Overall) | 81.8% ✎ | ▂··▂█▂▄▂▂·▃▂ beats 11/12 | 2.9% |
$ where to run it
Through an API
api · fal.ai · $0.053/image · fal.ai ↗ · price sourcefal.ai openai/gpt-image-2, 1024×1024 medium output price. Input tokens and any billed text output are additional; not OpenAI direct endpoint tariff. Product: Text-to-image endpoint. API model: openai/gpt-image-2. Settings: quality=medium, width=1024, height=1024, n=1. Price basis: image output component. Currency: USD.
$ how the score was worked out · LMArena leaderboard dataset 30%, Verifiable Visual Rewards Tra… 29%, InfoEdit: Probing Global Layo… 19% and 8 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