ZMIME
Image model comparison · 2 models · Updated Nov 2, 2026

GPT Image 2.5 Flare vs Qwen-Image 2.1

Preferred for text to image: GPT Image 2.5 Flare · Preferred for image editing: GPT Image 2.5 Flare · Newest: Qwen-Image 2.1. Quality scores are blind human-preference votes; specs and prices come from each lab’s listings.

  1. OpenAI

    GPT Image 2.5 Flare

    Released Sep 8, 2026New

    • Text to image
    • Image editing
    • Arena1,482
    • From—
    • Max size—
    Full specs Remove GPT Image 2.5 Flare

  2. Alibaba (Qwen)

    Qwen-Image 2.1

    Released Sep 14, 2026New

    • Text to image
    • Image editing
    • Arena1,362
    • From—
    • Max size—
    Full specs Remove Qwen-Image 2.1

  3. Add a model

    Make it a three-way comparison.

01 — At a glance

Where they differ

Quality is the arena score from blind, side-by-side human votes. Price, length and resolution come from each lab’s own listings.

  • Preferred for text to imageGPT Image 2.5 FlareGPT Image 2.5 Flare 1,401 (#2) · Qwen-Image 2.1 1,225 (#19) — blind human votes
  • Preferred for image editingGPT Image 2.5 FlareGPT Image 2.5 Flare 1,482 (#2) · Qwen-Image 2.1 1,362 (#20) — blind human votes
  • NewestQwen-Image 2.1GPT Image 2.5 Flare Sep 8, 2026 · Qwen-Image 2.1 Sep 14, 2026
  • Widest inputsSame inputsGPT Image 2.5 Flare: Text prompts, Reference images · Qwen-Image 2.1: Text prompts, Reference images
  • Image editingBoth edit imagesTakes a reference image as input: GPT Image 2.5 Flare yes · Qwen-Image 2.1 yes
  • Self-hostingQwen-Image 2.1Downloadable weights (qwen-research)
  • Most API hostsGPT Image 2.5 FlareGPT Image 2.5 Flare: 4 · Qwen-Image 2.1: 1
02 — Side by side

Every spec in one table

Dashes mean the data is not published.

GPT Image 2.5 Flare vs Qwen-Image 2.1 specifications side by side
SpecificationGPT Image 2.5 FlareOpenAIQwen-Image 2.1Alibaba (Qwen)
Quality (blind human votes)
Text to image arenaElo-style score · higher is better1,401 ±5#2 of 82 · 39,218 votes (best)1,225 ±6#19 of 82 · 19,604 votes
Image editing arenaElo-style score · higher is better1,482 ±5#2 of 61 · 87,941 votes (best)1,362 ±6#20 of 61 · 39,412 votes
Price
How it is billedBilled per token: $30 per million output image tokens.—
Output
ProducesImagesImages
Images per requestUp to 10—
Inputs
Text promptsYesYes
Reference images (editing)YesYes
Max reference images16—
Access
WeightsProprietaryOpenqwen-research
API hosts4 (best)1
ReleasedSep 8, 2026Sep 14, 2026
03 — Questions

Which should you choose?

Which is better, GPT Image 2.5 Flare or Qwen-Image 2.1?

For text to image, voters prefer GPT Image 2.5 Flare (GPT Image 2.5 Flare 1,401 (#2) · Qwen-Image 2.1 1,225 (#19)). For image editing, voters prefer GPT Image 2.5 Flare (GPT Image 2.5 Flare 1,482 (#2) · Qwen-Image 2.1 1,362 (#20)). Arena scores come from blind, side-by-side human votes on the same prompts; when the gap is inside both models’ error margins we call it a tie. Test your own prompts before you commit.

What is the difference between GPT Image 2.5 Flare and Qwen-Image 2.1?

GPT Image 2.5 Flare is an image model from OpenAI (Sep 8, 2026), taking text prompts and reference images and producing images. Qwen-Image 2.1 is an image model from Alibaba (Qwen) (Sep 14, 2026), taking text prompts and reference images and producing images.

Which is newer?

GPT Image 2.5 Flare Sep 8, 2026; Qwen-Image 2.1 Sep 14, 2026.

Which can edit an existing image?

Takes a reference image as input: GPT Image 2.5 Flare yes; Qwen-Image 2.1 yes.

Can I run any of these on my own hardware?

Qwen-Image 2.1. Downloadable weights (qwen-research).

04 — Keep comparing

Related comparisons