Model Details
Qwen3.5 122B A10B
Quick answer: Qwen3.5 122B A10B is Alibaba's September 2025 MoE model scoring 63.8% BrowseComp, 47.5% HLE, 86.6% GPQA Diamond, 86.7% MMLU-Pro, and 72% SWE-Bench Verified. Apache 2.0 with 122B total / 10B active parameters.
At a Glance
Where Qwen3.5 122B A10B leads
- 47.5% HLE — among the top open-weight HLE scores; near GLM-5.2 (54.7%) and Gemini 3 Flash (43.5%)
- 63.8% BrowseComp — strong web research
- 86.6% GPQA Diamond — excellent graduate-level science
- 86.7% MMLU-Pro — strong academic knowledge
- Apache 2.0 — fully open
- 10B active params: efficient MoE inference
Where it lags
- 72% SWE-Bench Verified — below Kimi K2.6 (80.2%) and HY3 (78%)
- September 2025: superseded by more recent models on some metrics
Best for: Open-weight frontier reasoning with 47.5% HLE; web research at 63.8% BrowseComp; academic knowledge tasks; teams needing Apache 2.0 models with Qwen 3.5 generation capabilities.
What Qwen3.5 122B A10B Is
Qwen3.5 122B A10B is the September 2025 flagship in Alibaba's Qwen3.5 generation — succeeding Qwen3 (April/May 2025). As a 122B/10B MoE, inference activates only 10B parameters per token while the full 122B capacity is available for specialised knowledge routing.
The 47.5% HLE score is notably high for an open-weight model — near Gemini 3.1 Pro (46.4%) and GLM-5.2 (54.7%) while being Apache 2.0 open. This makes it one of the strongest openly-available models for academic and research reasoning.
Specifications
| Field | Value |
|---|---|
| Organization | Alibaba |
| License | Apache 2.0 |
| HuggingFace | Qwen/Qwen3.5-122B-A10B |
| Release date | September 2025 |
| Parameters | 122B total / 10B active (MoE) |
| Modality | Text only |
| Context window | 128K tokens |
Pricing
Open weights under Apache 2.0 — self-host. Also via Alibaba Cloud DashScope.
Public Benchmark Scores
| Benchmark | Score | Source | Date |
|---|---|---|---|
| BrowseComp | 63.8% | Benchgen evaluation | 2025-09 |
| GPQA Diamond | 86.6% | Benchgen evaluation | 2025-09 |
| Humanity's Last Exam | 47.5% | Benchgen evaluation | 2025-09 |
| MMLU-Pro | 86.7% | Benchgen evaluation | 2025-09 |
| SWE-Bench Verified | 72% | Benchgen evaluation | 2025-09 |
Qwen3.5 122B A10B vs Alternatives
| Model | HLE | BrowseComp | MMLU-Pro | License |
|---|---|---|---|---|
| Qwen3.5 122B A10B | 47.5% | 63.8% | 86.7% | Apache 2.0 |
| Kimi K2 Thinking 0905 | 51.0% | 60.2% | 84.6% | Apache 2.0 |
| Kimi K2.6 | 36.4% | 83.2% | — | Apache 2.0 |
| Gemini 3 Flash | 43.5% | — | — | Proprietary |
Qwen3.5 122B A10B vs Kimi K2 Thinking 0905: higher HLE (47.5% vs 51%), higher MMLU-Pro (86.7% vs 84.6%), higher BrowseComp (63.8% vs 60.2%). Very close competition; both Apache 2.0.
Related Models & Benchmarks
- Compare: Kimi K2 Thinking 0905 · Kimi K2.6 · Qwen3 235B A22B
- Benchmarks: BrowseComp · Humanity's Last Exam · GPQA Diamond
Frequently Asked Questions
What is Qwen3.5 122B A10B?
Alibaba's September 2025 MoE model scoring 47.5% HLE, 63.8% BrowseComp, 86.6% GPQA Diamond, 86.7% MMLU-Pro, 72% SWE-Bench. Apache 2.0 with 122B/10B active params.How does Qwen3.5 122B compare to Qwen3 235B A22B?
Qwen3 235B A22B (April 2025) excels on Arena Hard (95.6%) and code tasks. Qwen3.5 122B A10B (September 2025) offers higher HLE (47.5%) and more recent training. For research/reasoning: Qwen3.5 122B. For instruction quality: Qwen3 235B.Specs from Alibaba's Qwen3.5 122B A10B release (September 2025) and Benchgen evaluations. Last updated 2026-07-24.
Benchmark Leaderboards
This model isn’t on any benchmark leaderboard yet.