Qwen3.8 Flash Next
qwen
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
ReasoningTool CallingAttachmentsOpen Weightsqwen-community-1.0
Context Window
262K
Max Output
131K
Temperature
Yes
Open Weights
Yes
Knowledge Cutoff
N/A
Released
2026-08-27
Last Updated
2026-08-27
License
qwen-community-1.0
Modalities
Input:TextImageVideo
→Output:Text
Benchmarks
| Benchmark | Score | Metric | Harness |
|---|---|---|---|
| DeepSWE | 58.7 | score | mini-swe-agent |
| SWE-Bench Pro | 62.5 | score | Claude Code |
| SWE-Bench Multilingual | 81 | score | mini-swe-agent |
| NL2Repo | 48.1 | score | Claude Code |
| 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 | Claude Code |
| 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 | - |
Available Providers (0)
Tidak ada provider yang terdaftar untuk model ini.