GPT Image 2.5 is live — OpenAI's newest image model, targeted edits that leave the rest of the frame alone
All Posts
Qwen Image 3 vs GPT Image 2: Text, Editing, and Price
2026/10/07

Qwen Image 3 vs GPT Image 2: Text, Editing, and Price

Qwen Image 3 vs GPT Image 2 (and Nano Banana 2): arena Elo for text, layout and editing, output limits, official per-image prices, and which to pick per job.

Qwen Image 3 vs GPT Image 2 comes down to a trade between quality and cost. On Artificial Analysis, where people vote blind between pairs of images, GPT Image 2 at high quality scores 1172 Elo for text-to-image. Qwen-Image-3.0-Pro scores 1088[1]. At list price, though, a square GPT Image 2 image at high quality costs $0.211 from OpenAI[2], while Alibaba lists a 1K Qwen-Image-3.0-Pro image at $0.04[3].

This article compares the two models on the four things people ask about: text and layout, editing, output size and price. Nano Banana 2 is included as a third reference point because many readers compare Qwen Image 3 against it too. Each section ends with a plain recommendation.

TL;DR

  • Overall preference: GPT Image 2 (high) 1172, Nano Banana 2 1126, Qwen-Image-3.0-Pro 1088, Qwen-Image-3.0 1075 on the Artificial Analysis text-to-image board[1].
  • Text rendering: GPT Image 2 (high) ranks first in the text-rendering category with 1207. Qwen-Image-3.0-Pro has 1090[4].
  • Layout: Qwen-Image-3.0-Pro (1181) places ahead of Nano Banana 2 (1154) but behind GPT Image 2 (1222)[5].
  • Price: Alibaba lists Qwen-Image-3.0 at $0.03 per image at both 1K and 2K[3]. Google lists Nano Banana 2 at $0.067 per 1K image[6].
  • Output size: Qwen Image 3.0 tops out at 2048×2048 total pixels[7]. GPT Image 2 goes up to 3840 pixels on the long edge[8].

Qwen Image 3 vs GPT Image 2 at a glance

Qwen Image 3.0 (Standard / Pro)GPT Image 2Nano Banana 2
ReleasedJuly 21, 2026[9]April 21, 2026[10]GA May 28, 2026[11]
Prompt lengthUp to 4.5k tokens[9]Not statedNot stated
Max output2048×2048 total pixels[7]3840 px long edge, up to 8,294,400 pixels[8]4K[6]
Reference images1–3[7]Multiple, plus a mask[8]Up to 14[12]
Masked edits (inpainting)No mask input[7]Mask image[13]Described in the prompt[12]
Transparent backgroundNot documentedPreview since Aug 20, 2026[10]Not documented
Official list price, 1K$0.03 / $0.04 per image[3]$0.006 low, $0.053 medium, $0.211 high (1024×1024)[2]$0.067[6]
Arena Elo, text-to-image1075 / 1088[1]1172 (high)[1]1126[1]
Arena Elo, editing1056 / 1077[14]1125 (high)[14]1110[14]

Prices are the vendors' own list prices, checked on October 7, 2026. Alibaba's figures are for the Singapore region. Elo scores are a snapshot from the same day and will move as votes come in.

What the arena scores say about text and layout

Qwen sells Qwen Image 3.0 on text. Its launch post says the model renders "text as small as 10px", supports 12 languages natively and composes newspapers, exam papers and 3×3 infographic grids from a single prompt[9]. Alibaba's API reference lists both tiers with the same input and output limits, and describes Standard as the tier that "balances quality and speed"[7].

The blind votes do not put Qwen first on text, though. Artificial Analysis ranks models by Elo from pairwise human preference votes[1], and its text-rendering category has GPT Image 2 (high) at number one with 1207. Nano Banana 2 has 1109, Qwen-Image-3.0-Pro 1090 and Qwen-Image-3.0 1078[4]. The confidence intervals for Nano Banana 2 (1087–1131) and Qwen-Image-3.0-Pro (1064–1116) overlap, so those two are close to a tie on text.

Layout is where Qwen does best. In the layout category Qwen-Image-3.0-Pro scores 1181 and ranks sixth. GPT Image 2 (high) has 1222, and Nano Banana 2 has 1154[5]. That matches what Qwen advertises: lots of panels and captions arranged on one page. OpenAI itself lists "precise text placement" and "layout-sensitive compositions" among the limitations of GPT Image models[8]. On these votes, though, GPT Image 2 still ranks above Qwen on layout.

Our reading of Qwen Image 3 vs GPT Image 2 on text: if the quality of the text decides the job and budget does not, GPT Image 2 at high quality has the strongest evidence. If you need dense, multi-panel pages at a low price per image, Qwen-Image-3.0-Pro gets you close at a fraction of the cost.

Editing: references, masks and transparency

All three models edit from a text instruction plus reference images, but the controls are different.

  • Qwen Image 3.0 takes 1 to 3 reference images and returns 1 to 6 outputs per request. Alibaba's API reference has no mask parameter, so the edit is described in the prompt[7].
  • GPT Image 2 supports inpainting[13]. You upload a mask and the edit is limited to that area. When there are several input images, OpenAI applies the mask to the first one[8]. Transparent backgrounds have been in preview for gpt-image-2 since August 20, 2026[10].
  • Nano Banana 2 accepts up to 14 reference images, the most of the three. Google's guide describes masking as something you define in the conversation rather than with a mask file[12].

Votes on the editing board follow the same order as text-to-image: GPT Image 2 (high) 1125, Nano Banana 2 1110, Qwen-Image-3.0-Pro 1077, Qwen-Image-3.0 1056[14]. For product cutouts, local fixes or anything that needs an alpha channel, GPT Image 2 is the only one of the three with the controls. For combining many reference images, Nano Banana 2 accepts the most.

Output size and resolution

Qwen Image 3.0 outputs between 512×512 and 2048×2048 total pixels[7]. GPT Image 2 accepts any size whose edges are multiples of 16, with a long edge up to 3840 pixels, a ratio no wider than 3:1 and up to 8,294,400 pixels in total, so 3840x2160 is a valid size[8]. Nano Banana 2 is priced up to 4K[6].

If you print large posters or need 4K files straight from the model, Qwen Image 3.0 is the only one of the three that cannot do it.

What each image costs at list price

The three vendors bill in different ways:

  • Alibaba (Qwen Image 3.0): per image. Qwen-Image-3.0 is $0.03 at 1K and at 2K. Qwen-Image-3.0-Pro is $0.04 at 1K and $0.075 at 2K. Each input image adds $0.003. These are Singapore list prices[3].
  • OpenAI (GPT Image 2): by tokens, at $30 per million image output tokens[2]. OpenAI's estimate for a 1024×1024 image is $0.006 at low, $0.053 at medium and $0.211 at high quality[8].
  • Google (Nano Banana 2): $0.067 at 1K, $0.101 at 2K and $0.151 at 4K[6].

Taking these list prices together: the quality setting that Artificial Analysis tested for GPT Image 2 (high) costs about five times a 1K Qwen-Image-3.0-Pro image ($0.211 against $0.04). At medium quality the gap narrows to about 1.3 times ($0.053 against $0.04). There is no arena score for GPT Image 2 at medium, so that comparison is about price only.

On reAPI, all three are called through the same /api/v1/images/generations endpoint. Qwen Image 3.0 is billed per delivered image. Standard costs the same at 1K and 2K, Pro charges twice as much at 2K as at 1K, and reference images are billed once per request[15]. GPT Image 2 comes as two model ids. gpt-image-2 is a flat price per resolution. gpt-image-2-official is metered by tokens and adds quality tiers, masks and transparent backgrounds[16]. Current reAPI rates are on the Qwen Image 3.0, GPT Image 2 and Nano Banana 2 model pages. For a full price breakdown of Qwen, see Qwen Image 3.0 API pricing.

Which model to pick for which job

JobPickWhy
Menus, newspapers, exam papers, multi-panel infographics on a budgetQwen-Image-3.0-ProRanks sixth on layout[5]; $0.04 at 1K on Alibaba's list[3]
High-volume general images where 2K mattersQwen-Image-3.0Same list price at 1K and 2K[3]
Text accuracy is the main requirementGPT Image 2 (high)First in the text-rendering category[4]
Masked edits, cutouts, transparent PNGsGPT Image 2Inpainting and transparent backgrounds[13][10]
4K outputGPT Image 2 or Nano Banana 2Qwen Image 3.0 stops at 2048×2048[7]
Many reference images in one editNano Banana 2Up to 14 references[12]

This Qwen Image 3 vs GPT Image 2 table is our editorial judgment, based on the scores and specifications cited above. Run a few of your own prompts through each model before you move production traffic.

FAQ

Qwen image 3 vs Nano Banana 2?

Nano Banana 2 has more votes behind it overall (1126 against 1088 for Qwen-Image-3.0-Pro)[1] and in editing[14]. Qwen-Image-3.0-Pro places higher on layout (1181 against 1154), although the intervals overlap[5]. Qwen lists for less ($0.04 against $0.067 at 1K)[3][6]. Google deprecated Nano Banana 2 on October 6, 2026 in favour of Nano Banana 2.1 and has not announced a shutdown date[11].

Qwen-Image-3.0-Pro vs GLM Image?

They are different kinds of product. GLM-Image is an open-weight model under the MIT licence, built from a 9B autoregressive generator and a 7B diffusion decoder[17]. On Artificial Analysis it scores 891 for text-to-image and 785 for editing. Qwen-Image-3.0-Pro scores 1088 and 1077[1][14]. Pick GLM-Image if you need to run the model on your own hardware, and Qwen-Image-3.0-Pro if output quality matters more.

Qwen Image 3 benchmark

Qwen's launch post includes no benchmark table[9]. The independent numbers come from Artificial Analysis: Qwen-Image-3.0-Pro has 1088 and Qwen-Image-3.0 has 1075 for text-to-image, ranking 14th and 16th on October 7, 2026[1].

Qwen Image 3.0 text rendering capabilities

Qwen says Qwen Image 3.0 renders text down to 10px, covers 12 languages and takes prompts up to 4.5k tokens, which is enough to describe a full page[9]. In the blind text-rendering votes, Qwen-Image-3.0-Pro scores 1090, behind GPT Image 2 (high) at 1207 and close to Nano Banana 2 at 1109[4].

Qwen Image 3 Pro edit

Pass 1 to 3 reference image URLs together with an instruction, and Qwen-Image-3.0-Pro edits instead of generating from scratch[7]. There is no mask input. On the editing board the Pro tier scores 1077 and the standard tier 1056[14]. Parameters are in the Qwen Image 3.0 API docs.

GPT image 2.5 vs Nano Banana pro vs Nano Banana 2 vs Qwen image 3 vs Seedream 5.0 pro, what do you think is better?

On the Artificial Analysis text-to-image board on October 7, 2026, the order is GPT Image 2.5 Sunburst (max) 1198, GPT Image 2.5 Flare (max) 1191, Nano Banana 2 1126, Nano Banana Pro 1102, Qwen-Image-3.0-Pro 1088, Seedream 5.0 Pro 1081 and Qwen-Image-3.0 1075[1]. OpenAI released GPT Image 2.5 on September 8, 2026[10]. It is on reAPI as GPT Image 2.5.

Choosing between Qwen Image 3 and GPT Image 2

GPT Image 2 wins on the evidence: it scores higher on human preference, on text, on layout and on editing, and it is the only one of the two with masks, transparency and 4K output. Qwen Image 3.0 is the choice when cost per image decides. Its Pro tier is within reach of Nano Banana 2 on text and ahead of it on layout, at a lower list price. Qwen Image 3 vs GPT Image 2 is therefore a budget decision first. If you are unsure, try the same three prompts on both: Qwen Image 3.0 on the model page and GPT Image 2 on its model page. Whether Qwen Image 3.0 can be run locally is covered in Is Qwen Image 3 open source?.

References

  1. Artificial Analysis. Text to Image leaderboard (AA-Image-T2I v2.0). Retrieved October 2026 from artificialanalysis.ai/image/leaderboard/text-to-image
  2. OpenAI. API pricing. Retrieved October 2026 from developers.openai.com/api/docs/pricing
  3. Alibaba Cloud. Model Studio: model inference pricing. Retrieved October 2026 from alibabacloud.com/help/en/model-studio/model-pricing
  4. Artificial Analysis. Text to Image leaderboard, Text Rendering category. Retrieved October 2026 from artificialanalysis.ai/image/leaderboard/text-to-image?category=text-rendering
  5. Artificial Analysis. Text to Image leaderboard, Layout category. Retrieved October 2026 from artificialanalysis.ai/image/leaderboard/text-to-image?category=layout
  6. Google. Gemini Developer API pricing. Retrieved October 2026 from ai.google.dev/gemini-api/docs/pricing
  7. Alibaba Cloud. Qwen Image Generation and Editing API reference. Retrieved October 2026 from alibabacloud.com/help/en/model-studio/qwen-image-generation-and-editing-api-reference
  8. OpenAI. Image generation guide. Retrieved October 2026 from developers.openai.com/api/docs/guides/image-generation
  9. Qwen Team. Qwen-Image-3.0: Rich Content, Authentic Details, Deep Knowledge. Retrieved October 2026 from qwen.ai/blog?id=qwen-image-3.0
  10. OpenAI. API changelog. Retrieved October 2026 from developers.openai.com/api/docs/changelog
  11. Google. Gemini API release notes. Retrieved October 2026 from ai.google.dev/gemini-api/docs/changelog
  12. Google. Nano Banana image generation. Retrieved October 2026 from ai.google.dev/gemini-api/docs/image-generation
  13. OpenAI. GPT-Image-2 model page. Retrieved October 2026 from developers.openai.com/api/docs/models/gpt-image-2
  14. Artificial Analysis. Image Editing leaderboard (AA-Image-Editing v2.0). Retrieved October 2026 from artificialanalysis.ai/image/leaderboard/editing
  15. reAPI. Qwen Image 3.0 API reference. Retrieved October 2026 from reapi.ai/docs/qwen-image-3
  16. reAPI. GPT Image 2 stable API reference. Retrieved October 2026 from reapi.ai/docs/gpt-image-2-stable
  17. Z.ai. GLM-Image model card. Retrieved October 2026 from huggingface.co/zai-org/GLM-Image
All Posts

More Posts