When was Qwen3.8-27B released?

Qwen3.8-27B was released by Qwen on Friday, Aug 14 2026.

Who made Qwen3.8-27B?

Qwen3.8-27B was built by Qwen. Alibaba's AI lab, building the Qwen family. The most prolific publisher of open-weight models of any major lab, alongside a proprietary Max and Plus tier sold through Alibaba Cloud.

What benchmark scores did Qwen3.8-27B get?

Qwen3.8-27B reports 16 tracked benchmark scores — BullshitBench v2: 78%; SWE-Bench Pro: 61.7%; DeepSWE 1.1: 42.2%; NL2Repo-Bench: 42.3%; QwenSWEBench: 79%; LiveCodeBench: 90.3%; Terminal-Bench 2.1: 73%; JobBench: 33.4%; CoWorkBench: 70.7%; Humanity's Last Exam (no tools): 30.8%; Humanity's Last Exam (with tools): 30.8%; GPQA Diamond: 89.2%; IFBench: 79.5%; OSWorld-Verified: 84.3%; Agent's Last Exam (pass@1): 20.4%; Agent's Last Exam (score): 42.9%. Scores are the figures published at release by Qwen. It holds the best score among all models tracked here on QwenSWEBench.

What is the context window of Qwen3.8-27B?

Qwen3.8-27B has a context window of 262k. That is the maximum amount of input plus output the model can hold in a single request.

How many parameters does Qwen3.8-27B have?

Qwen3.8-27B is reported at 27B parameters.

Is Qwen3.8-27B open source?

Partly. Qwen3.8-27B is an open-weight model: the trained weights are free to download and run locally or on your own infrastructure, but the training data and code are not fully released and the license may restrict some commercial uses. It is not open source in the strict sense.

What came before and after Qwen3.8-27B?

Qwen's previous tracked release was Qwen3.8-Max on Aug 3 2026, 11 days earlier. It was followed by Qwen3.8-Flash-Next on Aug 26 2026.

Qwen3.8-27BOpen Weight

Released Aug 14 2026

Qwen3.8-27B is an AI model released by Qwen on Friday, Aug 14 2026, 11 days after Qwen3.8-Max. It is an open-weight model — the trained weights are available to download and run. It is a 27B parameter model with a 262k token context window. Benchmark results (shown below) cover QwenSWEBench, BullshitBench v2, SWE-Bench Pro, DeepSWE 1.1, NL2Repo-Bench, LiveCodeBench, and 8 more.

Available from


Darkbloom	$0.15	$2.00	256K	fp4
Parasail	$0.24	$2.20	256K	fp8
AkashML	$0.25	$2.20	256K	fp8
Chutes	$0.32	$2.50	256K	fp8
Reka	$0.214	$2.55	256K	fp8
Ionstream	$0.35	$2.55	256K	fp8
Io Net	$0.30	$2.80	64K	fp8
Phala	$0.30	$3.00	256K	—
CoreWeave	$0.40	$3.00	256K	fp8
Novita	$0.42	$3.00	1M	—
Venice	$0.45	$3.20	256K	fp8

USD per 1M tokens

Benchmarks

Coding

QwenSWEBench

79%

SWE-Bench Pro

61.7%

#11 of 23

DeepSWE 1.1

42.2%

#20 of 25

NL2Repo-Bench

42.3%

#5 of 5

LiveCodeBench

90.3%

#5 of 12

Terminal & CLI

Terminal-Bench 2.1

73%

#22 of 29

Agentic & tool use

JobBench

33.4%

#7 of 8

CoWorkBench

70.7%

#4 of 4

OSWorld-Verified

84.3%

via BenchLM

#4 of 28

Agent's Last Exam

20.4%

pass@1

42.9%

score

pass@1

#6 of 6

score

#2 of 2

Reasoning & science

Humanity's Last Exam

30.8%

no tools

30.8%

with tools

via BenchLM

no tools

#15 of 21

with tools

#34 of 41

GPQA Diamond

89.2%

#19 of 58

IFBench

79.5%

#2 of 2

Robustness

BullshitBench v2

78%

#9 of 74

Source: BenchLM, retrieved 31 August 2026. Every other score here is the figure the lab published at launch.

About

Qwen3.8-27B, released August 14, 2026, brought Alibaba's Qwen3.8 generation to a size intended for local deployment: a 27B-parameter dense, natively multimodal model with downloadable weights under Apache 2.0. Its 262K-token native context window could be extended to 1M through YaRN, giving builders much of the long-context reach of Qwen3.8-Max without the flagship's 2.4-trillion-parameter footprint or hosted-only multimodal product.

The launch table concentrated on coding and office agents. Qwen3.8-27B scored 61.7% on SWE-Bench Pro, 42.2% on DeepSWE 1.1 and 79.0% on QwenSWEBench at release, all sizable gains over the earlier Qwen3.6-27B. Its 70.7% on CoWorkBench also edged the 68.2% Alibaba reported for Opus 4.6 Max, while 90.3% on LiveCodeBench v6 and 89.2% on GPQA Diamond put the compact model close to the larger proprietary systems in Qwen's comparison. Twelve days later it became the baseline the lab measured its Qwen4 architecture preview against, when Qwen3.8-Flash-Next shipped and beat it on almost every row of the same table.

Compare Qwen3.8-27B with

Qwen3.8-27B

Suggested comparisons

Qwen3.8-27BvsQwen3.8-Max-0902 Qwen3.8-27BvsGPT-6 Astra Qwen3.8-27BvsClaude Fable 5.1 Qwen3.8-27BvsGemini 3.8 Flash Qwen3.8-27BvsMuse Spark 1.3 Qwen3.8-27BvsGrok 4.6 Qwen3.8-27BvsDeepSeek-V4-Pro-0813 Qwen3.8-27BvsMistral Medium 3.5 Qwen3.8-27BvsKimi K3 Qwen3.8-27BvsGLM-5.3-Flash Qwen3.8-27BvsNemotron 3.5 Lightning

Frequently asked questions

: Qwen3.8-27B was released by Qwen on Friday, Aug 14 2026.
: Qwen3.8-27B was built by Qwen. Alibaba's AI lab, building the Qwen family. The most prolific publisher of open-weight models of any major lab, alongside a proprietary Max and Plus tier sold through Alibaba Cloud.
: Qwen3.8-27B reports 16 tracked benchmark scores — BullshitBench v2: 78%; SWE-Bench Pro: 61.7%; DeepSWE 1.1: 42.2%; NL2Repo-Bench: 42.3%; QwenSWEBench: 79%; LiveCodeBench: 90.3%; Terminal-Bench 2.1: 73%; JobBench: 33.4%; CoWorkBench: 70.7%; Humanity's Last Exam (no tools): 30.8%; Humanity's Last Exam (with tools): 30.8%; GPQA Diamond: 89.2%; IFBench: 79.5%; OSWorld-Verified: 84.3%; Agent's Last Exam (pass@1): 20.4%; Agent's Last Exam (score): 42.9%. Scores are the figures published at release by Qwen. It holds the best score among all models tracked here on QwenSWEBench.
: Qwen3.8-27B has a context window of 262k. That is the maximum amount of input plus output the model can hold in a single request.
: Qwen3.8-27B is reported at 27B parameters.
: Partly. Qwen3.8-27B is an open-weight model: the trained weights are free to download and run locally or on your own infrastructure, but the training data and code are not fully released and the license may restrict some commercial uses. It is not open source in the strict sense.
: Qwen's previous tracked release was Qwen3.8-Max on Aug 3 2026, 11 days earlier. It was followed by Qwen3.8-Flash-Next on Aug 26 2026.