Qwen 3.5 122B A10B: benchmark results
Alibaba's open Qwen 3.5 122B A10B MoE (10B active). Provider: Alibaba. Released 2026-02-24. Access: Open.
Aggregate Rating 1710 ± 4, rank #173 of 1656 rated models, from 1551 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| CUDAHercules - Single-Kernel Pass Rate (Self-Refine) | 90 | Pass rate (%; share of the 20 architecture-general Class 1 s | 100 |
| CircuitReason-1k - Mixed and General | 84.5 | Accuracy (%; exact problem-level correctness on the 258 mixe | 100 |
| CircuitReason-1k - Multi-Step Reasoning | 89.51 | Accuracy (%; exact problem-level correctness on the 429 mult | 100 |
| Equational Theories - Hard (Low-or-none reasoning) | 59.14 | Strict F1 (%) | 100 |
| Equational Theories - Normal (Low-or-none reasoning) | 97.53 | Strict F1 (%) | 100 |
| FullStackBench en | 62.6 | Score (imported) | 100 |
| LLM Stats (Codeforces - Qwen3.5 Unresolved Source Fraction) | 0.85 | Reported source fraction (metric unresolved) | 100 |
| LLM Stats (VideoMME w/o sub.) | 83.9 | Score (%) | 100 |
| MATH-MC Level 5 | 99.92 | Accuracy (%) | 100 |
| MERA Multi - CommonVideoQA | 74.3 | Score (%) | 100 |
| MERA Multi - LabTabVQA | 82.3 | Score (%) | 100 |
| MERA Multi - SchoolScienceVQA | 77.6 | Score (%) | 100 |
Interactive version: theaggregate.ai/model?slug=qwen-3-5-122b-a10b · How It Works · Data refreshed daily, snapshot 2026-10-20.