Qwen3.8-27B
Alibaba · Released 17 Aug 2026 · qwen3.8-27b
$0.50
Input price / 1M tokens
#14 of 61, cheapest first
$3.00
Output price / 1M tokens
#19 of 61, cheapest first
1M
Context window
#29 of 71
46
Output tokens / second
#54 of 61
73%
Terminal-Bench 2.1
#15 of 17
61.7%
SWE-bench Pro
#7 of 16
84.3%
OSWorld-Verified
#2 of 8
89.2%
GPQA Diamond
#17 of 27
Summary
Qwen3.8-27B is an open-weight model from Alibaba, released on 17 Aug 2026.
At $0.50 input and $3.00 output per million tokens, it is #14 of 61 on input price, cheapest first.
Its 1M-token context window ranks #29 of 71.
Measured output speed is 46 tokens per second, #54 of 61.
Dense 27B native vision-language model (images and video), open weights under Apache-2.0 with no extra commercial conditions. Weights support 262,144 tokens natively, extensible to 1M; the hosted API serves 1M by default.
Benchmarks · 6 reported
| Benchmark | Score | Bar | Rank | Source |
|---|---|---|---|---|
| Terminal-Bench 2.1Terminus harness | 73% | #15 of 17 | Qwen3.8-27B model cardSelf-reported | |
| SWE-bench Pro | 61.7% | #7 of 16 | Qwen3.8-27B model cardSelf-reported | |
| OSWorld-Verified | 84.3% | #2 of 8 | Qwen3.8-27B model cardSelf-reported | |
| GPQA Diamond | 89.2% | #17 of 27 | Qwen3.8-27B model cardSelf-reported | |
| Humanity’s Last Exam (no tools)no tools | 30.8% | #15 of 23 | Qwen3.8-27B model cardSelf-reported | |
| LiveCodeBench | 90.3% | #1 of 7 | Qwen3.8-27B 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 Qwen3.8-27B, and the closest scores.
Terminal-Bench 2.1
#15 of 17Terminal tasks, earlier set
- Qwen3.8-27B73
- Qwen3.7-Max74.5
- GLM-5.281
- Gemini 3.5 Flash76.2
- Kimi K388.3
SWE-bench Pro
#7 of 16Harder multi-file engineering tasks
- Qwen3.8-27B61.7
- Qwen3.7-Max60.6
- GLM-5.262.1
- GPT-5.558.6
- Gemini 3.5 Flash53.9
OSWorld-Verified
#2 of 8Using a desktop computer
- Qwen3.8-27B84.3
- GPT-5.578.7
- Gemini 3.5 Flash78.4
- Kimi K384.8
GPQA Diamond
#17 of 27Graduate-level science questions
- Qwen3.8-27B89.2
- Qwen3.7-Max92.4
- GLM-5.291.2
- GPT-5.593.6
- Kimi K393.5
Compare Qwen3.8-27B side by side
Pre-filled with the two closest models. Swap any of them.
Sources
- price: alibabacloud.com/help/en/model-studio/model-pricing, checked 2 Oct 2026
- context, max output, cache prices, modalities: qwencloud.com/models/qwen3.8-27b, checked 2 Oct 2026
- API release date: alibabacloud.com/help/en/model-studio/newly-released-models, checked 2 Oct 2026
- license, parameters, benchmarks: huggingface.co/Qwen/Qwen3.8-27B, checked 2 Oct 2026
- output speed: artificialanalysis.ai/models/qwen3-8-27b, checked 2 Oct 2026
Building on Qwen3.8-27B?
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.