Qwen3.8-Max
Alibaba · Released 2 Aug 2026 · qwen3.8-max
$2.00
Input price / 1M tokens
#36 of 61, cheapest first
$6.00
Output price / 1M tokens
#32 of 61, cheapest first
1M
Context window
#29 of 71
39
Output tokens / second
#57 of 61
86.6%
Terminal-Bench 2.1
#8 of 17
67.7%
SWE-bench Pro
#1 of 16
92.6%
GPQA Diamond
#8 of 27
43.6%
Humanity’s Last Exam (no tools)
#5 of 23
Summary
Qwen3.8-Max is an open-weight model from Alibaba, released on 2 Aug 2026.
At $2.00 input and $6.00 output per million tokens, it is #36 of 61 on input price, cheapest first.
Its 1M-token context window ranks #29 of 71.
Measured output speed is 39 tokens per second, #57 of 61.
Alibaba's current flagship, built on the open-weight Qwen3.8-2.4T-A95B (2.4T total, 95B active MoE). The API version adds vision input, non-thinking mode and 1M context by default. Weights license allows commercial use, but products over 100M MAU or $20M monthly revenue must show the model name, and Model-as-a-Service or AI coding/office assistant businesses with over $50M revenue in 12 months need a separate license from Qwen.
Benchmarks · 6 reported
| Benchmark | Score | Bar | Rank | Source |
|---|---|---|---|---|
| Terminal-Bench 2.1 | 86.6% | #8 of 17 | Qwen3.8-2.4T-A95B model cardSelf-reported | |
| SWE-bench Pro | 67.7% | #1 of 16 | Qwen3.8-2.4T-A95B model cardSelf-reported | |
| GPQA Diamond | 92.6% | #8 of 27 | Qwen3.8-2.4T-A95B model cardSelf-reported | |
| Humanity’s Last Exam (no tools)no tools | 43.6% | #5 of 23 | Qwen3.8-2.4T-A95B model cardSelf-reported | |
| Humanity’s Last Exam (with tools)with tools | 56.2% | #10 of 16 | Qwen3.8-2.4T-A95B model cardSelf-reported | |
| Toolathlon VerifiedPass@1 | 72.5% | – | Qwen3.8-2.4T-A95B model cardSelf-reported |
Self-reported means the lab ran the test itself. A rank appears only where several labs report the same version of a benchmark.
Similar models
Models with the most benchmarks in common with Qwen3.8-Max, and the closest scores.
Terminal-Bench 2.1
#8 of 17Terminal tasks, earlier set
- Qwen3.8-Max86.6
- GLM-5.281
- Kimi K388.3
- DeepSeek V4 Pro87.9
- DeepSeek V4.1 Flash90.6
GPQA Diamond
#8 of 27Graduate-level science questions
- Qwen3.8-Max92.6
- GLM-5.291.2
- Kimi K393.5
- DeepSeek V4 Pro92.4
- GPT-5.593.6
- DeepSeek V4.1 Flash90.9
Humanity’s Last Exam (no tools)
#5 of 23Expert questions, no search or code
- Qwen3.8-Max43.6
- GLM-5.240.5
- Kimi K343.5
- DeepSeek V4 Pro42.7
- GPT-5.541.4
- DeepSeek V4.1 Flash36.8
Compare Qwen3.8-Max side by side
Pre-filled with the two closest models. Swap any of them.
Sources
- price (Singapore international, Global), batch/cache rules: alibabacloud.com/help/en/model-studio/model-pricing, checked 2 Oct 2026
- cache prices, context, max output, modalities: qwencloud.com/models/qwen3.8-max, checked 2 Oct 2026
- release date (Singapore): alibabacloud.com/help/en/model-studio/newly-released-models, checked 2 Oct 2026
- parameters, license, benchmarks: huggingface.co/Qwen/Qwen3.8-2.4T-A95B, checked 2 Oct 2026
- output speed: artificialanalysis.ai/models/qwen3-8-max, checked 2 Oct 2026
Building on Qwen3.8-Max?
Get the architecture right before the bill arrives.
We size caching, routing and fallbacks for your workload on a free call, and tell you where a cheaper model is good enough.