Qwen3 VL 235B A22B
alibaba/qwen3-vl-235b-a22b-instruct- Alibaba
- Family · Qwen3 VL
- Released 23 Sep 2025
- Open weights
- Text · Image → Text
- Tools · JSON · Vision
Not on OSMEXO yet
Spec & list price
Context256K
Max output32K
Input / 1M$0.21
Output / 1M$1.909× input
Scores · standing
Intelligence9.9#248/489
Usage
Tokens / day4.5B▲ 26% 7d
Usage rank#116
Requests / day1.0M
Tool calls / req0.01
Per request · 30 days
Avg input3.5K0.2× median
Avg output1570.2× median
Cache hit43%0.6× median
Tool-call failures2.8%1.5× median
Consumption
Peak 4.7B 29 OctLow 2.2B 18 OctRank today #116
News & trends
1 mentions this week ▼ 50%Qwen3 VL family · videos, posts, articles · 30 days
Trending · 14 days · Qwen3 VL family
- FLUX.3 Image Klein (5.6B + Qwen3-VL-4B-Instruct) across 15 complex prompts: Macro, Typography, Fluid physics, and 0.5x Selfies, Workflows Include5 upvotes
- FLUX.3 Image Klein (5.6B + Qwen3-VL-4B-Instruct) across 15 complex prompts: Macro, Typography, Fluid physics, and 0.5x Selfies, Workflows Include5 upvotes
News · latest from Alibaba
- Memdoor: A team of 5 Qwen3.8-27B models beats Fable 5 on LiveCodeBench Hardgithub.com · 31 Oct
- Qwen3.8-Flash Now Available in the EU on Alibaba Cloudtwitter.com · 31 Oct
- Alibaba's Qwen3.8-Flash-Next model tried to upload a file to Beijing in a testnotebookcheck.net · 27 Oct
- Running Qwen3.8-Flash-Next NVFP4 at 184 tok/s on a single RTX 6000 Promeshapplied.com · 26 Oct
Benchmarks
bar = share of the top score · Benchmark data: Artificial AnalysisSpeed · median across 3 providers, latest day
Output speed#295/33430 tok/s2.2× slower than median
Fastest provider33 tok/sbest of 3 providers
First token#2311.58 s1.5× longer than median
#24 cheapest of 41 with a similar score
Value
Intelligence Index
Blended price per 1M tokens (log)
Family progress
Intelligence Index
- Family releases11.9
Technical details
- Tokenizer
- Qwen3
- Knowledge cutoff
- 2025-03-31
- License
- Apache 2.0
- Parameters
- 236B
- Moderation
- none
- Downloads · likes
- 298.3K · 429
Request parameters 17
- tools
- tool_choice
- response_format
- structured_outputs
- temperature
- top_p
- top_k
- min_p
- max_tokens
- stop
+7 more
- seed
- frequency_penalty
- logit_bias
- logprobs
- presence_penalty
- repetition_penalty
- top_logprobs
Interesting facts
- Less output room than similar models (32.8K against a 65.5K median).
- The most used of 6 Qwen3 VL models by tokens last month.
When to use
Best for
- Image understanding in chat apps
- Visual document and screenshot analysis
- Open-weight multimodal assistants
- Science and math questions with images
Think twice for
- Video or audio input
- Image generation
- Long-context reasoning over huge inputs
- High-volume simple chat (smaller options are cheaper)
Research papers
- WOVEN: Weaving Visual World Modeling into Multimodal LLMs8 Oct
- Not Just Reason, Not Just Scan: Reinforcement Learning for Proactive Scientific Error Verification over Academic Paper27 Aug
- MemeMind: Reference-Guided Trace Construction for Offline Context Optimization10 Aug
- CapRL++: Unified Reinforcement Learning with Verifiable Rewards for Dense Image and Video Captioning8 Jun
- UniEditBench: A Unified and Cost-Effective Benchmark for Image and Video Editing via Distilled MLLMs17 Apr
- Scientific Graphics Program Synthesis via Dual Self-Consistency Reinforcement Learning7 Apr
- Internalized Reasoning for Long-Context Visual Document Understanding31 Mar
- Ostrakon-VL: Towards Domain-Expert MLLM for Food-Service and Retail Stores29 Jan
Same family
Qwen3 VL- 13.4
- 9.5
- 8.4
- 11.9
- 7.0
- 8.2
- 5.7
Alternatives
other vendors · similar capability- 8.5
- 11.6
- 13.1
- 6.7
Want Qwen3 VL 235B A22B on OSMEXO?
Sign in to request it. Requests are reviewed; a request does not guarantee availability.