Qwen 3.5 397B
alibaba/qwen3.5-397b-a17b- Alibaba
- Family · Qwen3
- Released 16 Feb 2026
- Open weights
- Text · Image · Video → Text
- Tools · JSON · Vision
Spec & list price
Context256K
Max output236K
Input / 1M$0.50
Output / 1M$3.607.2× input
Scores · standing
Intelligence21.4#129/489
Coding48.2#72/196
Usage
Tokens / day1.2B▲ 3.2% 7d
Usage rank#189
Requests / day129.4K
Tool calls / req0.27
Per request · 30 days
Avg input8.1K0.5× median
Avg output1.3K1.9× median
Cache hit37%0.5× median
Reasoning share73%1.3× median
Tool-call failures3.9%2.1× median
Consumption
Peak 13.0B 29 SepLow 630.6M 25 OctRank today #189
News & trends
184 mentions this week ▲ 36%Qwen3 family · videos, posts, articles · 30 days
Trending · 14 days · Qwen3 family
- Memdoor: A team of 5 Qwen3.8-27B models beats Fable 5 on LiveCodeBench Hard453 upvotes
- Memdoor: A team of 5 Qwen3.8-27B models beats Fable 5 on LiveCodeBench Hard416 points
- Neuron Explorer for Qwen3.5-4B: See every neuron at work, make targeted edits26 points
- New Card for Qwen3.8 27b and MiniMax H3632 upvotes
- Qwen 3.8 27B vs Strata Coder - Local LLM battle75.2K views
- DavidAU Qwen3.8 27B Turbo Brilliance 50X-Reasoning... tested - Local LLM setup87.7K views
News · latest from Qwen3
- Memdoor: A team of 5 Qwen3.8-27B models beats Fable 5 on LiveCodeBench Hardgithub.com · 31 Oct
- Neuron Explorer for Qwen3.5-4B: See every neuron at work, make targeted editsgithub.com · 26 Oct
- How to train your own Jev for $17Together AI · 23 Sep
Benchmarks
bar = share of the top score · Benchmark data: Artificial AnalysisSpeed · median across 10 providers, latest day
Output speed#232/33449 tok/s1.3× slower than median
Fastest provider64 tok/sbest of 10 providers
First token#1891.22 s1.2× longer than median
First answer token#321.74 sthinking included
#23 cheapest of 29 with a similar score
Value
Intelligence Index
Blended price per 1M tokens (log)
Family progress
Intelligence Index
- Family releases33.7
Technical details
- Tokenizer
- Qwen3
- License
- Apache 2.0
- Parameters
- 403B
- Moderation
- none
- Downloads · likes
- 980.3K · 1.6K
Request parameters 19
- tools
- tool_choice
- response_format
- structured_outputs
- reasoning
- include_reasoning
- temperature
- top_p
- top_k
- min_p
+9 more
- max_tokens
- stop
- seed
- frequency_penalty
- logit_bias
- logprobs
- presence_penalty
- repetition_penalty
- top_logprobs
Interesting facts
- 2.2× the output room of similar models (235.9K against a 106.5K median).
- Weekdays average 5.2B tokens a day against 3.0B on weekends.
When to use
Best for
- Agentic tool calling and instruction following
- Image and video understanding
- Hard science question answering
- Long document reasoning with self-hosting
Think twice for
- Low-cost bulk classification
- Heavy terminal automation versus top coding models
- Latency-sensitive simple chat
Research papers
- HybridResearcher: Bridging the Structured-Unstructured Boundary for Deep Research Agents17 Oct
- ZoomCap: Evidence-Routed Visual Zooming for Recoverable Visual Information Transfer15 Oct
- Replay the Curvature: Accurate and Scalable NVFP4 Quantization for Large Language Model Inference29 Sep
- Teaching LLMs to Update Beliefs for Efficient Long-Horizon Interaction26 Jul
- Medical thinking with multiple images14 Apr
Same family
Qwen3- 7.2
- 6.0
- 6.9
- 6.1
- 5.2
- 13.1
Alternatives
other vendors · similar capability- 26.3
- 21.8
- T25.0
- 21.5
Using Qwen 3.5 397B at volume?
Sign in to ask for a volume discount. You will continue in the OSMEXO dashboard.