AlibabaQwen3 32B
Best configuration ranks #380 of 610 on the BenchLeader Index at 45.9 ±5.9. Last measured 2 Sept 2026. Released 1 Apr 2025.
- Blended price
- $0.502/M
- $0.287 in · $1.15 out
- Output speed
- 105 tok/s
- First answer
- 2.53 s
- first token 1.01 s
- Context
- 131k
- Overall index46
- Reasoning50
- Coding46
- Agents & tools59
- Maths53
- Instruction following35
- Human preference52
- Long context25
- Composite38
Reasoning-effort configurations
The same model behaves differently depending on how much it is allowed to think. Each row is one setting, scored only on the benchmarks that were run at that setting. “Default” means the publisher did not say which setting was used.
| Effort | Index | Rank | Speed | First answer | Chat reply cost | Categories |
|---|---|---|---|---|---|---|
| no reasoning | – | – | 105 tok/s | 2.53 s | $0.0005 | Maths 34 · Reasoning 43 |
| thinking | 41.0 | #497 | 104 tok/s | 22 s | $0.0003 | Agents & tools 36 · Composite 38 · Instruction following 39 · Knowledge 40 · Long context 25 |
| defaultbest | 45.9 | #380 | 105 tok/s | 2.53 s | $0.0005 | Agents & tools 59 · Coding 46 · Composite 38 · Human preference 52 · Instruction following 35 · Long context 25 · Maths 53 · Reasoning 50 |
Benchmark results
One column per reasoning effort. Rank is among every configuration of every model on that benchmark. Hover a score for the run it came from.
Reasoning
| Benchmark | no reasoning | thinking | default | Source |
|---|---|---|---|---|
| GPQA Diamond | 54.1%#191 | – | 65.7%#166 | Epoch AI Benchmarking Hub |
| LMArena Hard Prompts | – | – | 1367#177 | LMArena |
| GPQA Diamond (AA)not in index | – | 66.8%#313 | 53.5%#391 | Artificial Analysis |
| Humanity's Last Exam (AA)not in index | – | 7.4%#294 | 4.1%#453 | Artificial Analysis |
| Kagi LLM Benchmark | 48.7%#88 | – | – | Kagi LLM Benchmark |
Coding
| Benchmark | no reasoning | thinking | default | Source |
|---|---|---|---|---|
| SciCode | – | – | 35.4%#133 | SciCode |
| LMArena Coding | – | – | 1407#168 | LMArena |
| SciCode (AA)not in index | – | 36.0%#136 | – | Artificial Analysis |
| Aider Polyglot | – | – | 40.0%#29 | Aider polyglot leaderboard |
Agents & tools
| Benchmark | no reasoning | thinking | default | Source |
|---|---|---|---|---|
| Terminal-Bench Hard | – | 3.0%#298 | – | Artificial Analysis |
| τ²-Bench Telecom (AA)not in index | – | 29.8%#253 | – | Artificial Analysis |
| BFCL Overall | – | – | 48.7%#25 | Berkeley Function Calling Leaderboard |
Maths
| Benchmark | no reasoning | thinking | default | Source |
|---|---|---|---|---|
| OTIS Mock AIME | 23.1%#198 | – | 66.9%#133 | Epoch AI Benchmarking Hub |
Knowledge
| Benchmark | no reasoning | thinking | default | Source |
|---|---|---|---|---|
| AA-Omniscience | – | -50.4#316 | – | Artificial Analysis |
Instruction following
| Benchmark | no reasoning | thinking | default | Source |
|---|---|---|---|---|
| IFBench | – | 36.3%#287 | 31.5%#333 | Artificial Analysis |
Human preference
| Benchmark | no reasoning | thinking | default | Source |
|---|---|---|---|---|
| LMArena Text | – | – | 1347#183 | LMArena |
Long context
| Benchmark | no reasoning | thinking | default | Source |
|---|---|---|---|---|
| AA-LCR | – | 0.0%#419 | 0.0%#419 | Artificial Analysis |
Composite
| Benchmark | no reasoning | thinking | default | Source |
|---|---|---|---|---|
| Epoch Capabilities Indexnot in index | – | – | 138.5#104 | Epoch AI Benchmarking Hub |
| AA Intelligence Index | – | 7.2#408 | 7.3#402 | Artificial Analysis |
Where to run it
Every provider serving this model through OpenRouter, with throughput and first-token latency measured on live traffic over the last 30 minutes and each provider’s own price. Purple marks the best in each column.
| Provider | Speed | First token | Input $/M | Output $/M | Context | Quantisation |
|---|---|---|---|---|---|---|
| SiliconFlow | 30 tok/s | 1.45 s | $0.140 | $0.570 | 131k | fp8 |
| DeepInfra | 22 tok/s | 0.56 s | $0.080 | $0.280 | 41k | fp8 |
What a task costs
Estimates from list price, output speed and time to first answer for the best configuration. “With caching” assumes three-quarters of the input is served from the prompt cache. Reasoning tokens are not modelled.
| Workload | Tokens in / out | Cost | With caching | Time |
|---|---|---|---|---|
| Chat reply | 400 / 300 | $0.0005 | – | 5.4 s |
| Summarise a 30-page report | 12,000 / 600 | $0.0041 | – | 8.2 s |
| Code edit | 6,000 / 1,500 | $0.0034 | – | 16.8 s |
| Agentic coding session | 60,000 / 4,000 | $0.022 | – | 40.5 s |
| Structured extraction | 2,000 / 200 | $0.0008 | – | 4.4 s |
See also
Data as of 9 Sept 2026. Compare these configurations.