Alibaba
Qwen3.8 Flash Next
Open-weight experimental preview of the Qwen4 architecture: hybrid-attention MoE (125B total, 6B active) with vision encoder for coding, agent tasks, and image and video understanding
alibaba/qwen3.8-flash-next
Context
262.1K
Output
131.1K
Released
2026-08-27
Knowledge cutoff
—
Types
Reasoning
Tool use
Structured output
Attachments
Open weights
Provider pricing
No providers currently list this model.
Benchmarks
DeepSWE
58.7 · score
SWE-Bench Pro
62.5 · score
SWE-Bench Multilingual
81 · score
NL2Repo
48.1 · score
CoWorkBench
73.9 · score
JobBench
55.7 · score
Agents' Last Exam
24.3 · pass@1
Agents' Last Exam
51.2 · score
Toolathlon-Verified
73.5 · pass@1
IFBench
81.3 · score
GPQA Diamond
91.7 · score
Humanity's Last Exam
35.9 · score
LiveCodeBench
91.9 · score
ClawEval-MM
64.4 · pass@3
ClawEval-MM
60.4 · average score
RecreationBench
49.9 · score
AndroidWorld
84.5 · score
OSWorld
19.4 · binary completion rate
OSWorld
52.3 · partial score
Vision2Web
64 · score
ERQA
72.3 · score
LVBench
76.6 · score
RealWorldQA
88.5 · score
MathVision
90.6 · score
MathVision
95.7 · score
CharXiv Reasoning
84.6 · score
CharXiv Reasoning
90.6 · score