Qwen Image 3.0 Pro
TL;DR Highlight
Qwen-Image-3.0-Pro renders complex layouts like newspapers and menus in one pass
Who Should Read
A backend or product developer comparing cost and quality before adding an image generation API to a service
Core Mechanics
- Complex layouts in a single pass
- It accepts up to 4.5k input tokens.
- It claims dense layouts with images inside images, such as newspapers, storyboards, menus and exam papers, in one generation.
- Text as small as 10px
- It claims precise rendering of text down to about 10px.
- It also claims fine details like micro-expressions, pores and single hair strands close to real photos.
- 12 languages and many fonts
- It renders 12 languages natively and supports various fonts.
- It also claims realistic simulation of interfaces like web pages, games and live streams.
- Output costs $0.04 at 1K, $0.075 at 2K
- Pricing is per image: $0.04 for 1K output and $0.075 for 2K output.
- Image input is $0.003 per image at both 1K and 2K.
- Limits: 10 concurrent, 5 per minute
Community Reactions
- Layouts weaker than gpt-image-2, per one user
- A commenter used it for a week after launch and called it a big step up from the prior model.
- They said it is nowhere near gpt-image-2 for high-density UI design.
- With the same prompts for website mockups, text was fairly good but layout and aesthetics lagged.
- Arena score 1263 vs 1380
- A comment cited the arena.ai text-to-image leaderboard.
- It lists gpt-image-2 at 1380 and this model at 1263.
- Also available via OpenRouter
- A comment shared openrouter.ai/qwen/qwen-image-3-pro.
- Others complained about the official sign-up flow.
- Complaints about payment and ID requirements
- One user said the official Qwen site blocks the screen with a payment method popup.
- Another said sign-up asked for a driver's license photo.
- One suggested waiting until it is on OpenRouter.
How to Apply
- Compare against gpt-image-2 on your own prompts
- Today, pick 3 real prompts such as menus, banners or infographics and run the same text on both models.
- Score text accuracy and layout separately on a checklist.
- Estimate monthly budget per image
- Multiply expected monthly generations by $0.04 for 1K or $0.075 for 2K.
- Add $0.003 per reference image if you send inputs.
- Design calls around the 5 RPM limit
- Queue requests and rate-limit to 5 per minute.
- Send bulk jobs through the async queue (up to 200).
- Pick an access route and check sign-up friction
- Check whether the official sign-up requires a payment method.
- If that is a burden, test the same model through OpenRouter.
Try It
# Paste into a terminal (first: export DASHSCOPE_API_KEY=your_key)
curl --location 'https://ap-southeast-1.maas.qwencloudapi.com/api/v1/services/aigc/multimodal-generation/generation' \
--header 'Content-Type: application/json' \
--header "Authorization: Bearer $DASHSCOPE_API_KEY" \
--data '{"model":"qwen-image-3.0-pro","input":{"messages":[{"role":"user","content":[{"text":"{{your image description}}"}]}]},"parameters":{"prompt_extend":true}}'
# Next: generate the same description with gpt-image-2 and compare text and layout.
# There is a 5 requests-per-minute limit, so space out consecutive calls.Terminology
Related Papers
Mercury 3: Diffusion at scale
Scaling diffusion LLM 40x gives Mercury 3: 4–8x faster than speed-tier models, similar quality
Timestral: Mistral's time-series foundation models
A 209M time-series model ranks first on three benchmarks and runs on CPU.
DeepSeek V4.1 Flash 1017 (Max)
DeepSeek V4.1 Flash 1017 Max scores 45 on the Intelligence Index at 211 tok/s and $0.30 input.
If you don't run your own LLM proxy, how do you know whats going on?
Running your own LLM proxy shows which keys, models, and cache misses drive your usage.
DeepSeek: DeepSeek v4.1 Flash 1017
DeepSeek V4.1 Flash 1017: $0.05/M input, 1M context, built for agents
DeepSeek-v4.1-Flash-1017
DeepSeek V4.1 Flash minor update lifts coding and agent benchmarks while cutting run time and output length
Show HN: Llmspeed – feel how fast a local LLM runs before you buy the hardware