AlibabaQwen3.6 35B-A3B
Best configuration ranks #232 of 610 on the BenchLeader Index at 53.2 ±6.3. Last measured 14 Apr 2026. Released 14 Apr 2026.
- Blended price
- $0.557/M
- $0.248 in · $1.49 out
- Output speed
- 126 tok/s
- First answer
- 45 s
- first token 1.05 s
- Context
- 262k
- Overall index53
- Reasoning61
- Coding39
- Agents & tools45
- Maths45
- Knowledge54
- Instruction following63
- Multimodal60
- Long context62
- Composite57
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 | 46.0 | #375 | 139 tok/s | 2.00 s | $0.0008 | Agents & tools 56 · Coding 5 · Composite 48 · Instruction following 39 · Knowledge 36 · Long context 58 · Maths 44 · Multimodal 55 · Reasoning 62 |
| defaultbest | 53.2 | #232 | 126 tok/s | 45 s | $0.0005 | Agents & tools 45 · Coding 39 · Composite 57 · Instruction following 63 · Knowledge 54 · Long context 62 · Maths 45 · Multimodal 60 · Reasoning 61 |
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 | default | Source |
|---|---|---|---|
| GPQA Diamond | 84.8%#81 | 83.8%#87 | Epoch AI Benchmarking Hub |
| GPQA Diamond (AA)not in index | 81.7%#173 | 84.1%#148 | Artificial Analysis |
| Humanity's Last Exam (AA)not in index | 13.9%#204 | 22.2%#148 | Artificial Analysis |
Coding
| Benchmark | no reasoning | default | Source |
|---|---|---|---|
| SciCode | 1.3%#158 | 35.8%#129 | SciCode |
| WeirdML | – | 34.5%#124 | WeirdML |
| SciCode (AA)not in index | – | 36.6%#130 | Artificial Analysis |
Agents & tools
| Benchmark | no reasoning | default | Source |
|---|---|---|---|
| Terminal-Bench | – | 23.0%#52 | Terminal-Bench |
| Terminal-Bench Hard | 25.8%#132 | 34.9%#80 | Artificial Analysis |
| τ²-Bench Telecom (AA)not in index | 85.1%#92 | 95.3%#24 | Artificial Analysis |
Maths
| Benchmark | no reasoning | default | Source |
|---|---|---|---|
| FrontierMath Tiers 1–3 | 20.4%#72 | 17.5%#83 | Epoch AI Benchmarking Hub |
| OTIS Mock AIME | 68.9%#126 | 86.7%#71 | Epoch AI Benchmarking Hub |
Knowledge
| Benchmark | no reasoning | default | Source |
|---|---|---|---|
| AA-Omniscience | -60.0#389 | -22.2#187 | Artificial Analysis |
Instruction following
| Benchmark | no reasoning | default | Source |
|---|---|---|---|
| IFBench | 36.2%#289 | 64.3%#108 | Artificial Analysis |
Multimodal
| Benchmark | no reasoning | default | Source |
|---|---|---|---|
| MMMU-Pro | 71.0%#110 | 75.0%#76 | Artificial Analysis |
Long context
| Benchmark | no reasoning | default | Source |
|---|---|---|---|
| AA-LCR | 64.3%#197 | 71.7%#144 | Artificial Analysis |
Composite
| Benchmark | no reasoning | default | Source |
|---|---|---|---|
| Epoch Capabilities Indexnot in index | – | 143.9#79 | Epoch AI Benchmarking Hub |
| AA Intelligence Index | 15.2#220 | 22.3#150 | 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 |
|---|---|---|---|---|---|---|
| CoreWeave | 138 tok/s | 0.23 s | $0.250 | $1.25 | 262k | fp8 |
| io.net | 114 tok/s | 0.93 s | $0.133 | $0.941 | 262k | fp8 |
| Phala | 94 tok/s | 2.38 s | $0.200 | $1.27 | 262k | – |
| SiliconFlow | 71 tok/s | 1.91 s | $0.200 | $1.60 | 262k | fp8 |
| AkashML | 68 tok/s | 0.57 s | $0.100 | $0.900 | 262k | fp8 |
| Venice | 48 tok/s | 0.65 s | $0.100 | $1.00 | 256k | fp8 |
| Darkbloom | 45 tok/s | 3.72 s | $0.050 | $0.700 | 262k | fp4 |
| AtlasCloud | 44 tok/s | 1.17 s | $0.186 | $1.11 | 262k | fp8 |
| Parasail | 39 tok/s | 0.58 s | $0.150 | $1.00 | 262k | fp8 |
| DeepInfra | 23 tok/s | 1.75 s | $0.100 | $0.950 | 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. Reasoning tokens are not modelled.
| Workload | Tokens in / out | Cost | With caching | Time |
|---|---|---|---|---|
| Chat reply | 400 / 300 | $0.0005 | – | 47.3 s |
| Summarise a 30-page report | 12,000 / 600 | $0.0039 | – | 49.7 s |
| Code edit | 6,000 / 1,500 | $0.0037 | – | 56.8 s |
| Agentic coding session | 60,000 / 4,000 | $0.021 | – | 1.3 min |
| Structured extraction | 2,000 / 200 | $0.0008 | – | 46.5 s |
See also
Data as of 9 Sept 2026. Compare these configurations.