Prism StudiosModel Leaderboard
Live · data checked Oct 18, 2026, 6:10 AM ET

October 2026 · Image

The best AI image models right now

Best overall

GPT Image 2.5 (Sunburst)

OpenAI

GPT Image 2.5 Sunburst is number 1 on the image arena. Our own tests back the family: its predecessor, GPT Image 2 (high), had the best craft in our comic-page test, and GPT Image 2.5 is what we use for a real person's likeness after another tool changed the face in 2 of 6 images.

Our testA comic page with a recurring characterSep 4, 2026Arena#1 of 104Image arena, as GPT Image 2.5 Sunburst (max)Cost$0.21 per imagelisted by Artificial Analysis
Why it won, with the evidence

Best value

Nano Banana 2.1

#4 on the image arena · $0.0336 per image (price listed by Artificial Analysis)

Would win about 43 of 100 blind matchups against #1.

The rule: the cheapest model that would still win at least 4 of every 10 blind matchups against #1 (within 70 Elo points). A formula, not an opinion, so no one can buy their way onto it.

Best for the job

If you needUseEvidenceWhat to know
A real person's likenessGPT Image 2.5 (Sunburst)Our test, Sep 25, 2026Give it 2 reference photos of the person (a headshot and a turnaround). Only do this with the person's permission.
Cheap, high-volume stillsMuse ImageOur test, Oct 3, 2026Our test, Sep 4, 2026One cent an image. It beat Nano Banana 2 at turning cartoon characters into photoreal actors. It refuses violent prompts.
Mature or violent scenes (comics, horror)Qwen Image 3Our test, Sep 4, 2026The only model that cleared all four bars on our violent comic page. Muse refused the prompt and Flux Kontext Pro returned a black frame.
Print-size pagesGemini 3.1 Flash Image (Nano Banana 2)Our test, Sep 4, 2026Ask for 4K or you get a small default image. The 4K page came back at 3392 by 5056 pixels.

Watch out

  • Flux Kontext Pro On a violent prompt it returned a fully black image with no error message. You only find out when you look.

See it for yourself

Our test

Cartoon cast to photoreal actors

Which cheap image model turns cartoon character designs into believable actors and stays on model?

How: Six cartoon character references from our show Pillbox, each converted to a photoreal actor by two models with the same prompt.

Muse Image result
Muse Image
Nano Banana 2 result
Nano Banana 2
Muse Image result
Muse Image
Nano Banana 2 result
Nano Banana 2
  • Muse Image

    Stayed closer to each design and used the plain gray backdrop the prompt asked for, every time. $0.01 an image.

  • Nano Banana 2

    Kept the pharmacy background on 2 of 5 characters even though the prompt asked for a plain gray backdrop, and changed one character's ethnicity. About $0.067 an image.

Limits: Both models added glasses that are not in some of the designs. One prompt, five characters shown here.

The public arenas this week

Image arena (AA-Image-T2I v2.1) · You give it a text prompt, it makes the picture. Top 10 · Artificial Analysis · read Oct 18, 2026, 6:10 AM ETThe score is an Elo rating, like chess: people compare two outputs without knowing which model made them, and winners gain points. The thin line is the margin of error. When two lines overlap, the models are effectively tied. Pink is our best overall pick. Linked names have their own evidence page.
  1. 1 GPT Image 2.5 Sunburst (max)OpenAI 1220$0.21 per image
  2. 2 GPT Image 2.5 Flare (max)OpenAI 1207$0.21 per image
  3. 3 GPT Image 2 (high)OpenAI 1186$0.21 per image
  4. 4 Nano Banana 2.1Google 1174$0.0336 per image
  5. 5 Grok Imagine Image 2.0SpaceXAI 1167$0.06 per image
  6. 6 MAI-Image-2.6Microsoft AI 1166$0.0389 per image
  7. 7 Nano Banana 2 (Gemini 3.1 Flash Image)Google 1140$0.067 per image
  8. 8 Muse ImageMeta 1127$0.01 per image
  9. 9 GPT Image 1.5 (high)OpenAI 1123$0.13 per image
  10. 10 MAI-Image-2.6-FlashMicrosoft AI 1119$0.0195 per image