Model Details
Qwen3 Next 80B A3B Thinking
Quick answer: Qwen3 Next 80B A3B Thinking is Alibaba's April 2026 thinking MoE model scoring 82.7% MMLU-Pro, 72.0% BFCL-v3, and 62.3% Arena Hard v2. Apache 2.0 — 80B/3B active.
At a Glance
Where Qwen3 Next 80B A3B Thinking leads
- 82.7% MMLU-Pro — best MMLU-Pro in Qwen3 Next 80B family
- 72.0% BFCL-v3 — strong function calling
- Apache 2.0 — fully open
- 3B active: low inference cost
Where it lags
- 62.3% Arena Hard — significantly lower than Instruct (82.7%)
Best for: MMLU-Pro knowledge + BFCL function calling at 3B active cost; reasoning-heavy tasks.
What Qwen3 Next 80B A3B Thinking Is
The thinking (chain-of-thought reasoning) variant of Qwen3 Next 80B A3B. Thinking raises MMLU-Pro to 82.7% (vs 80.6% instruct) and BFCL to 72.0% (vs 70.3%) but reduces Arena Hard significantly (62.3% vs 82.7%).
Specifications
| Field | Value |
|---|---|
| Organization | Alibaba |
| License | Apache 2.0 |
| HuggingFace | Qwen/Qwen3-Next-80B-A3B-Thinking |
| Release date | April 2026 |
| Parameters | 80B 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 |
|---|---|---|---|
| MMLU-Pro | 82.7% | Benchgen evaluation | 2026-04 |
| BFCL-v3 | 72.0% | Benchgen evaluation | 2026-04 |
| Arena Hard v2 | 62.3% | Benchgen evaluation | 2026-04 |
Qwen3 Next 80B Thinking vs Alternatives
| Model | MMLU-Pro | BFCL-v3 | Arena Hard | Type |
|---|---|---|---|---|
| Qwen3 Next 80B A3B Thinking | 82.7% | 72.0% | 62.3% | Thinking MoE |
| Qwen3 Next 80B A3B Instruct | 80.6% | 70.3% | 82.7% | Instruct MoE |
Thinking vs Instruct tradeoff: Thinking +2.1% MMLU-Pro, +1.7% BFCL, but -20.4% Arena Hard. Use Instruct for instruction following; Thinking for knowledge/reasoning.
Related Models & Benchmarks
- Compare: Qwen3 Next 80B A3B Instruct · Qwen3.6 35B A3B
- Benchmarks: MMLU-Pro · BFCL-v3 · Arena Hard v2
Frequently Asked Questions
What is Qwen3 Next 80B A3B Thinking?
Alibaba's April 2026 thinking MoE model scoring 82.7% MMLU-Pro, 72.0% BFCL-v3. Apache 2.0 — 80B/3B active.Specs from Alibaba's Qwen3 Next 80B A3B Thinking release (April 2026) and Benchgen evaluations. Last updated 2026-07-24.
Benchmark Leaderboards
This model isn’t on any benchmark leaderboard yet.