Qwen3.8-Flash-Next: benchmark results
Alibaba's open-weight Qwen3.8-Flash-Next (August 2026), an experimental preview of the Qwen4 architecture with a new hybrid attention design, a 125-billion-parameter MoE activating 6B parameters per token. Provider: Alibaba. Released 2026-08-26. Access: Open.
Aggregate Rating 1794 ± 11, rank #72 of 1656 rated models, from 451 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| BRIDGE Medical Leaderboard - CoT | 45.92 | Average Performance (%) | 100 |
| BRIDGE Medical Leaderboard - Few-Shot | 56.03 | Average Performance (%) | 100 |
| BRIDGE Medical Leaderboard - Zero-Shot | 46.95 | Average Performance (%) | 100 |
| Data Agent Benchmark (Built-in ReAct Agent) | 70.87 | Pass@1 (%) | 100 |
| ERQA | 72.3 | Score (%) | 100 |
| Jais 3 Model Cards - APEX-Agents (420 Text-Only Subset, Loop Agent) | 36.4 | Pass@1 (%) | 100 |
| Jais 3 Model Cards - Arabic File Processing (75% Threshold) | 82.7 | Accuracy (%) | 100 |
| Jais 3 Model Cards - Dialectal Arabic MMLU (IFM-audited) | 90.1 | Accuracy (%) | 100 |
| Jais 3 Model Cards - QIAS-2 (IFM-audited) | 87.8 | Accuracy (%) | 100 |
| Jais 3 Model Cards - SWE-bench Verified (Offline) | 80.8 | Pass@1 (%) | 100 |
| Jais 3 Model Cards - WildClawBench (English Text-Only Subset) | 59 | Mean Score@3 | 100 |
| LLM Stats (ClawEval-MM) | 60.4 | Score (%) | 100 |
Interactive version: theaggregate.ai/model?slug=qwen3-8-flash-next · How It Works · Data refreshed daily, snapshot 2026-10-20.