Model Details
Qwen3.6 35B A3B
Quick answer: Qwen3.6 35B A3B is Alibaba's December 2025 compact MoE model scoring 92.7% AIME 2026, 86.0% GPQA Diamond, 85.2% MMLU-Pro, and 75.3% MMMU-Pro. Apache 2.0 with 35B total / 3B active parameters.
At a Glance
Where Qwen3.6 35B A3B leads
- 92.7% AIME 2026 — excellent competition math at only 3B active params
- 86.0% GPQA Diamond — strong science reasoning
- 85.2% MMLU-Pro — strong academic knowledge
- 75.3% MMMU-Pro — solid multimodal understanding
- Apache 2.0 — fully open
- 3B active params: very low inference cost
Where it lags
- No SWE-Bench / BrowseComp data available
- December 2025: pre-dates Qwen3.5+ generation improvements
Best for: Compact open-weight deployments needing strong AIME/GPQA reasoning at 3B inference cost; cost-efficient STEM reasoning.
What Qwen3.6 35B A3B Is
Qwen3.6 35B A3B is the compact MoE model in the Qwen3.6 generation, released December 2025 alongside the Qwen3.6 Plus API model. "35B A3B" means 35B total with 3B active — making inference equivalent to a 3B dense model while maintaining 35B capacity.
At 3B active parameters, the 92.7% AIME 2026 is remarkable — showing Alibaba's training efficiency in the Qwen3.6 generation.
Specifications
| Field | Value |
|---|---|
| Organization | Alibaba |
| License | Apache 2.0 |
| HuggingFace | Qwen/Qwen3.6-35B-A3B |
| Release date | December 2025 |
| Parameters | 35B total / 3B active (MoE) |
| Modality | Text only |
Pricing
Open weights under Apache 2.0 — self-host at no cost.
Public Benchmark Scores
| Benchmark | Score | Source | Date |
|---|---|---|---|
| AIME 2026 | 92.7% | Benchgen evaluation | 2025-12 |
| GPQA Diamond | 86.0% | Benchgen evaluation | 2025-12 |
| MMLU-Pro | 85.2% | Benchgen evaluation | 2025-12 |
| MMMU-Pro | 75.3% | Benchgen evaluation | 2025-12 |
Qwen3.6 35B A3B vs Alternatives
| Model | AIME 2026 | GPQA Diamond | Active Params | License |
|---|---|---|---|---|
| Qwen3.6 35B A3B | 92.7% | 86.0% | 3B | Apache 2.0 |
| Qwen3 30B A3B | — | — | 3B | Apache 2.0 |
| Phi-4 Reasoning | — | 77.8% | 14B | Apache 2.0 |
Qwen3.6 35B A3B vs Qwen3 30B A3B: later generation (Dec 2025 vs May 2025) with AIME/GPQA coverage. For smallest-active-param reasoning: both are strong candidates at 3B active.
Related Models & Benchmarks
- Compare: Qwen3 30B A3B · Qwen3.6 Plus · Phi-4 Reasoning
- Benchmarks: AIME 2026 · GPQA Diamond · MMLU-Pro
Frequently Asked Questions
What is Qwen3.6 35B A3B?
Alibaba's December 2025 compact MoE model: 35B total / 3B active params. 92.7% AIME, 86.0% GPQA Diamond, 85.2% MMLU-Pro. Apache 2.0.Specs from Alibaba's Qwen3.6 35B A3B release (December 2025) and Benchgen evaluations. Last updated 2026-07-24.
Benchmark Leaderboards
This model isn’t on any benchmark leaderboard yet.