Qwen3 30b A3b Thinking · Qwen3-30B-A3B-Thinking-2507
Alibaba · 30 Jul 2025
benchmarks
Alibaba's Qwen3-30B-A3B-Thinking-2507 is an open-weight, text-only mixture-of-experts model with 30.5 billion total parameters and 3.3 billion activated per token. This dedicated thinking edition supports a native 262,144-token context window and targets reasoning-intensive mathematics, science and coding tasks.
Benchmark results
Every figure links to where it was published. Compare like with like — harnesses and prompt setups differ between vendors.
| Benchmark | Score | Source |
|---|---|---|
| AIMECompetition mathematics | 85%Jul 2025 | View |
| GPQA DiamondPhD-level science questions | 73.4%Jul 2025 | View |
Weighted mean of 2 verified benchmark results: AIME 85%, GPQA Diamond 73.4%.
Capabilities
- Dedicated thinking mode for multi-step reasoning.
- Mathematical and scientific question answering.
- Code generation and reasoning about programming problems.
- Native long-context text processing up to 262,144 tokens.
- Open weights for self-hosting and customization under the Apache 2.0 license.
What improved
- Native context increased from the original model's 32,768 tokens to 262,144 tokens.
- Vendor reports improved reasoning performance in mathematics, science and coding.
- Vendor reports better alignment with user preferences in subjective and open-ended tasks.
Known flaws & limitations
- Supports thinking mode only; it does not provide the original model's switchable non-thinking mode.
- Accepts text rather than native image, audio or video input.
- Extended reasoning can produce lengthy outputs, increasing latency and generated-token costs.
Risks, safety & security issues
- Not yet confirmed by public reporting.
Good fit for
- Multi-step mathematics and scientific problem solving.
- Programming assistance where extended reasoning is useful.
- Long-document analysis and text-based question answering.
- Self-hosted reasoning applications requiring access to model weights.
Poor fit for
- Applications requiring native image, audio or video understanding.
- Low-latency interactions that require disabling thinking.
- High-stakes decisions without independent validation and human oversight.
References
Profile last reviewed 10/27/2026 and re-checked automatically every few hours against the vendor's published documentation, papers and public leaderboards. Scores come from the cited benchmark results above — confirm limits and pricing against the provider's own documentation before building on them.