AlibabaQwen3 30B A3B 2507
Qwen3 30B A3B 2507 is an Alibaba open-weights reasoning model, released 30 Jul 2025. Its best configuration ranks #377 of 372 on the BenchLeader Index at 46.3 ±5.3, in the lower half. It scores highest in coding (58) and lowest in knowledge (33). At $0.150 per million tokens blended it is among the cheapest fifth of ranked models. Output speed of 141 tokens per second puts it in the fastest quarter, with a first answer in 1.9 s. It has been measured at 2 reasoning-effort settings; tables show the best-scoring one. Last measured 2 Sept 2026.
- Blended price
- $0.150/M
- $0.100 in · $0.300 out
- Output speed
- 141 tok/s
- measured by Artificial Analysis
- First answer
- 1.88 s
- first token 0.48 s
- Context
- 262k
- Overall index46
- Reasoning47
- Coding58
- Agents & tools48
- Maths52
- Knowledge33
- Instruction following37
- Human preference56
- Long context39
- Composite39
Versions
Alibaba has shipped 2 models under this name. Each is ranked on its own results; a newer version often has fewer results so far, which holds its index nearer the average until more arrive.
| Model | Released | Index | Rank |
|---|---|---|---|
| Qwen3 30B A3B 2507this page | 30 Jul 2025 | 46.3 | #377 |
| Qwen3 30B A3B | – | 42.4 | #471 |
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 |
|---|---|---|---|---|---|---|
| thinking | 45.0 | #409 | 149 tok/s | 16 s | $0.0008 | Agents & tools 38 · Coding 35 · Composite 41 · Instruction following 52 · Knowledge 37 · Long context 57 · Maths 38 · Reasoning 52 |
| defaultbest | 46.3 | #377 | 141 tok/s | 1.88 s | $0.0001 | Agents & tools 48 · Coding 58 · Composite 39 · Human preference 56 · Instruction following 37 · Knowledge 33 · Long context 39 · Maths 52 · Reasoning 47 |
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 | thinking | default | Source |
|---|---|---|---|
| GPQA Diamond | 70.1%#160 | 55.6%#192 | Epoch AI Benchmarking Hub |
| LMArena Hard Prompts | – | 1407#147 | LMArena |
| GPQA Diamond (AA)not in index | 70.7%#285 | 65.9%#327 | Artificial Analysis |
| Humanity's Last Exam (AA)not in index | 10.3%#258 | 6.9%#312 | Artificial Analysis |
| Kagi LLM Benchmark | 54.9%#67 | – | Kagi LLM Benchmark |
Coding
| Benchmark | thinking | default | Source |
|---|---|---|---|
| SciCode | 33.3%#140 | – | SciCode |
| LMArena Coding | – | 1439#139 | LMArena |
| SciCode (AA)not in index | 33.0%#141 | – | Artificial Analysis |
Agents & tools
| Benchmark | thinking | default | Source |
|---|---|---|---|
| Terminal-Bench Hard | 5.3%#268 | 6.1%#259 | Artificial Analysis |
| τ²-Bench Telecom (AA)not in index | 28.1%#261 | 10.2%#375 | Artificial Analysis |
| BFCL Overall | – | 41.4%#33 | Berkeley Function Calling Leaderboard |
Maths
| Benchmark | thinking | default | Source |
|---|---|---|---|
| OTIS Mock AIME | 70.3%#126 | 62.2%#146 | Epoch AI Benchmarking Hub |
| AIME 2026 | 88.3%#29 | – | MathArena |
| HMMT February 2026 | 78.8%#27 | – | MathArena |
| MathArena Apex | 0.5%#41 | – | MathArena |
Knowledge
| Benchmark | thinking | default | Source |
|---|---|---|---|
| AA-Omniscience | -56.1#373 | -66.2#435 | Artificial Analysis |
Instruction following
| Benchmark | thinking | default | Source |
|---|---|---|---|
| IFBench | 50.7%#169 | 33.1%#323 | Artificial Analysis |
Human preference
| Benchmark | thinking | default | Source |
|---|---|---|---|
| LMArena Text | – | 1382#152 | LMArena |
Long context
| Benchmark | thinking | default | Source |
|---|---|---|---|
| AA-LCR | 61.3%#214 | 26.3%#339 | Artificial Analysis |
Composite
| Benchmark | thinking | default | Source |
|---|---|---|---|
| AA Intelligence Index | 9.8#324 | 7.5#400 | 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 |
|---|---|---|---|---|---|---|
| Alibaba Cloud Int. | 69 tok/s | 0.32 s | $0.130 | $0.520 | 131k | – |
| StreamLake | 51 tok/s | 0.70 s | $0.048 | $0.193 | 128k | – |
| Nebius Token Factory | 23 tok/s | 0.48 s | $0.100 | $0.300 | 262k | fp8 |
| SiliconFlow | 16 tok/s | 1.46 s | $0.090 | $0.300 | 262k | fp8 |
Price history
Listed price per 1M tokens over time, as recorded by OpenRouter for the provider with the longest history.
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 at $0.010 per 1M. Reasoning tokens are not modelled.
| Workload | Tokens in / out | Cost | With caching | Time |
|---|---|---|---|---|
| Chat reply | 400 / 300 | $0.0001 | $0.0001 | 4.0 s |
| Summarise a 30-page report | 12,000 / 600 | $0.0014 | $0.0006 | 6.1 s |
| Code edit | 6,000 / 1,500 | $0.0011 | $0.0006 | 12.5 s |
| Agentic coding session | 60,000 / 4,000 | $0.0072 | $0.0032 | 30.3 s |
| Structured extraction | 2,000 / 200 | $0.0003 | $0.0001 | 3.3 s |
See also
Data as of 10 Sept 2026. Compare these configurations.