Update — October 13, 2026: OpenAI released GPT Image 2.5 on September 8. Its two models, Sunburst and Flare, now hold #1 and #2 on Arena’s text-to-image board, with GPT Image 2 third. Token prices did not change, so it is worth retesting your prompts on Flare before settling on GPT Image 2.
GPT Image 2 is OpenAI’s successor to DALL·E 3 and the model that set the bar for prompt understanding. When a brief is complex — many objects, specific spatial relationships, precise text — it follows instructions faithfully, and its conversational editing (“make the sky stormier”, “change the hat to red”) is among the easiest in the category.
Because it lives natively inside ChatGPT and the OpenAI API, it suits teams already in that ecosystem who want multi-step, iterative refinement rather than one-shot generation.
The trade-offs: at high quality it costs far more per image than FLUX.2, and its default aesthetic, while clean, lacks the distinctive character of Midjourney.
Where it stands in October 2026
| GPT Image 2 | GPT Image 2.5 Sunburst | GPT Image 2.5 Flare | |
|---|---|---|---|
| Arena text-to-image (Elo) | 1383 (#3) | 1425 (#1, preliminary) | 1398 (#2, preliminary) |
| Artificial Analysis text-to-image (Elo) | 1172 (high) | 1199 (max) | 1192 (max) |
| Artificial Analysis price per 1,000 images | $211.0 (high) | $210.7 (max) | $210.7 (max) |
| Token pricing (text in / image in / image out, per 1M) | $5 / $8 / $30 | $5 / $8 / $30 | $5 / $8 / $30 |
Arena and Artificial Analysis use different scales, so compare within a row. GPT Image 2’s Arena score rests on roughly 93,000 votes, against about 17,000 for each GPT Image 2.5 entry, so the newer scores may still move.
How it compares
- vs GPT Image 2.5: the same prices and API family, with higher Arena Elo, new
xhighandmaxquality tiers and tighter edit control. - vs Nano Banana 2.1: GPT Image 2 ranks higher on Arena (1383 vs 1328), while Nano Banana 2.1 is far cheaper per image.
- vs Midjourney v8.1: Midjourney is the pick for art-directed looks; GPT Image 2 is the pick for literal instruction following and edit chains.
- vs FLUX.2: FLUX.2 is much cheaper per image and can be self-hosted.
For other models in this category, see the image models directory and our Best AI Image Generator ranking.
Best for
- Conversational, iterative image editing
- ChatGPT-native workflows
- Marketing visuals needing precise instructions
Pros & cons
Strengths
- Strong instruction following for complex, multi-step prompts
- Easy conversational editing ("change the hat to red")
- Strong text rendering; native ChatGPT integration
- Still
Limitations
- Superseded by GPT Image 2.5, which ranks higher at the same token prices
- Expensive at high quality — about $0.21 per image on Artificial Analysis's default test, versus a few cents for FLUX.2
- Aesthetic is less distinctive than Midjourney
Frequently asked questions
Should I switch from GPT Image 2 to GPT Image 2.5?
Token prices are identical, and GPT Image 2.5 ranks higher on Arena's text-to-image board (Sunburst 1425 and Flare 1398 versus 1383 for GPT Image 2), so it is worth retesting your prompts on GPT Image 2.5 Flare first. Its rankings are still preliminary and output style shifts between versions, so compare on your own prompts before migrating.
How much does GPT Image 2 cost per image?
OpenAI bills tokens — $5 per 1M text input tokens, $8 per 1M image input tokens and $30 per 1M image output tokens — so the cost per image depends on size and quality. Artificial Analysis lists about $211 per 1,000 images (roughly $0.21 each) at high quality, 1024x1024.