Qwen3.8-27B

Name: Qwen3.8-27B
Author: Alibaba

Multimodal

Alibaba

Qwen3.8-27B is a dense 27.78-billion-parameter multimodal foundation model for coding, professional work, research, and long-horizon agents.

Key Specifications

Parameters

27.8B

Context

Release Date

August 14, 2026

Average Score

66.3%

Repository Model Weights Results Blog

Timeline

Key dates in the model's history

Announcement

August 14, 2026

Last Update

August 29, 2026

Today

August 30, 2026

Technical Specifications

Parameters

27.8B

Training Tokens

Knowledge Cutoff

Family

Capabilities

MultimodalZeroEval

Benchmark Results

Model performance metrics across various tests and benchmarks

Reasoning

Logical reasoning and analysis

GPQA

Diamond split • Self-reported

89.2%

Other Tests

Specialized benchmarks

Agents' Last Exam

Score • Self-reported

42.9%

AndroidWorld

• Self-reported

81.9%

BabyVision

With code interpreter • Self-reported

85.6%

CharXiv-R

With code interpreter • Self-reported

90.2%

ClawEval-MM

Average score across three trials • Self-reported

56.9%

CoWorkBench

• Self-reported

70.7%

DeepSWE 1.1

Claude Code, temp=1.0, top_p=0.95, 256K context • Self-reported

42.2%

ERQA

• Self-reported

65.5%

Humanity's Last Exam

Judged by GPT-4o • Self-reported

30.8%

IFBench

• Self-reported

79.5%

Job Bench

• Self-reported

33.4%

LiveCodeBench v6

• Self-reported

90.3%

MathVision

With code interpreter • Self-reported

94.6%

NL2Repo

Claude Code harness • Self-reported

42.3%

OmniDocBench 1.5

• Self-reported

91.1%

OSWorld-Verified

• Self-reported

84.3%

QwenSWEBench

Claude Code, avg@3, 8-hour timeout, max_tokens=32768, temp=1.0, 256K context • Self-reported

79.0%

RealWorldQA

• Self-reported

85.9%

RecreationBench

• Self-reported

47.1%

SWE-Bench Multimodal

Claude Code harness, public SWE-bench Multimodal dev split • Self-reported

38.6%

SWE-Bench Pro

Claude Code, temp=1.0, top_p=0.95, 256K context • Self-reported

61.7%

SWE-MM

Claude Code harness on the public SWE-bench Multimodal dev split • Self-reported

38.6%

Terminal-Bench 2.1

Terminus agent scaffold • Self-reported

73.0%

Vision2Web

Claude Code harness, judged by gpt-5.4-2026-03-05 • Self-reported

62.9%

WebArena-Verified

Official WebArena-Verified grader under the OSWorld scaffold • Self-reported

64.8%

License & Metadata

License

apache-2.0

Announcement Date

August 14, 2026

Last Updated

August 29, 2026

Compare Qwen3.8-27B

All comparisons

vs Qwen3.8 Flash Next vs Qwen3 VL 4B Instruct vs Qwen3.8 Max vs Qwen3.5-397B-A17B vs Qwen3.5 35B A3B vs Qwen3.5 122B A10B vs Qwen3.5 27B vs Qwen3.7 Max

Similar Models

All Models

Qwen3.6-35B-A3B

Alibaba

MM35.0B

Best score:0.9 (GPQA)

Released:Apr 2026

Qwen2-VL-72B-Instruct

Alibaba

MM73.4B

Released:Aug 2024

Qwen2.5 VL 72B Instruct

Alibaba

MM72.0B

Released:Jan 2025

Qwen3 VL 32B Thinking

Alibaba

MM33.0B

Released:Sep 2025

Qwen3.6-27B

Alibaba

MM27.8B

Released:Apr 2026

Price:$0.60/1M tokens

Qwen2.5 VL 32B Instruct

Alibaba

MM33.5B

Best score:0.9 (HumanEval)

Released:Feb 2025

QvQ-72B-Preview

Alibaba

MM73.4B

Released:Dec 2024

Qwen3.7-Plus

Alibaba

Best score:0.9 (GPQA)

Released:May 2026

Price:$0.32/1M tokens

Recommendations are based on similarity of characteristics: developer organization, multimodality, parameter size, and benchmark performance. Choose a model to compare or go to the full catalog to browse all available AI models.