AlibabaQwen3.5 397B-A17B
Best configuration ranks #120 of 610 on the BenchLeader Index at 58.6 ±4.3. Last measured 8 Sept 2026. Released 13 Feb 2026.
- Blended price
- $0.387/M
- $0.172 in · $1.03 out
- Output speed
- 81 tok/s
- First answer
- 41 s
- first token 1.30 s
- Context
- 262k
- Overall index59
- Reasoning63
- Coding52
- Agents & tools69
- Maths50
- Knowledge50
- Instruction following76
- Human preference63
- Multimodal62
- Long context65
- Composite53
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 | 52.5 | #247 | 76 tok/s | 2.06 s | $0.0013 | Agents & tools 65 · Composite 56 · Instruction following 52 · Knowledge 46 · Long context 58 · Maths 50 · Multimodal 37 · Reasoning 63 |
| thinking | – | – | 81 tok/s | 41 s | $0.0004 | Agents & tools 65 |
| defaultbest | 58.6 | #120 | 81 tok/s | 41 s | $0.0004 | Agents & tools 69 · Coding 52 · Composite 53 · Human preference 63 · Instruction following 76 · Knowledge 50 · Long context 65 · Maths 50 · Multimodal 62 · Reasoning 63 |
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 | Trend |
|---|---|---|---|---|---|
| GPQA Diamond | 86.4%#66 | – | 85.9%#72 | Epoch AI Benchmarking Hub | |
| LMArena Hard Prompts | – | – | 1463#74 | LMArena | |
| GPQA Diamond (AA)not in index | 86.1%#113 | – | 89.3%#79 | Artificial Analysis | |
| Humanity's Last Exam (AA)not in index | 19.8%#160 | – | 29.0%#112 | Artificial Analysis |
Coding
| Benchmark | no reasoning | thinking | default | Source | Trend |
|---|---|---|---|---|---|
| LMArena Coding | – | – | 1491#75 | LMArena | |
| LMArena WebDev | – | – | 1399#70 | LMArena | |
| SciCode (AA)not in index | – | – | 44.8%#94 | Artificial Analysis |
Agents & tools
| Benchmark | no reasoning | thinking | default | Source | Trend |
|---|---|---|---|---|---|
| Terminal-Bench Hard | 35.6%#72 | – | 40.9%#51 | Artificial Analysis | |
| τ²-Bench Telecom (AA)not in index | 83.9%#102 | – | 95.6%#20 | Artificial Analysis | |
| τ²-bench | – | 87.9%#1 | – | τ²-bench |
Maths
| Benchmark | no reasoning | thinking | default | Source | Trend |
|---|---|---|---|---|---|
| FrontierMath Tiers 1–3 | 31.2%#58 | – | 29.5%#60 | Epoch AI Benchmarking Hub | |
| OTIS Mock AIME | 82.2%#93 | – | 88.9%#61 | Epoch AI Benchmarking Hub | |
| AIME 2026 | – | – | 94.2%#20 | MathArena | |
| HMMT February 2026 | – | – | 87.9%#17 | MathArena |
Knowledge
| Benchmark | no reasoning | thinking | default | Source | Trend |
|---|---|---|---|---|---|
| AA-Omniscience | -37.9#250 | – | -30.8#218 | Artificial Analysis |
Instruction following
| Benchmark | no reasoning | thinking | default | Source | Trend |
|---|---|---|---|---|---|
| IFBench | 51.6%#163 | – | 78.8%#13 | Artificial Analysis |
Human preference
| Benchmark | no reasoning | thinking | default | Source | Trend |
|---|---|---|---|---|---|
| LMArena Text | – | – | 1441#79 | LMArena |
Multimodal
| Benchmark | no reasoning | thinking | default | Source | Trend |
|---|---|---|---|---|---|
| LMArena Vision | – | – | 1265#38 | LMArena | |
| MMMU-Pro | 52.7%#201 | – | 77.3%#59 | Artificial Analysis |
Long context
| Benchmark | no reasoning | thinking | default | Source | Trend |
|---|---|---|---|---|---|
| AA-LCR | 64.3%#197 | – | 77.3%#91 | Artificial Analysis |
Composite
| Benchmark | no reasoning | thinking | default | Source | Trend |
|---|---|---|---|---|---|
| Epoch Capabilities Indexnot in index | – | – | 147.0#55 | Epoch AI Benchmarking Hub | |
| AA Intelligence Index | 21.4#160 | – | 19.1#184 | 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 |
|---|---|---|---|---|---|---|
| GMICloud | 85 tok/s | 1.58 s | $0.600 | $3.60 | 262k | fp8 |
| StreamLake | 72 tok/s | 2.50 s | $0.600 | $3.60 | 256k | – |
| NovitaAI | 59 tok/s | 0.93 s | $0.600 | $3.60 | 262k | – |
| AtlasCloud | 53 tok/s | 1.24 s | $0.550 | $3.50 | 262k | fp8 |
| Parasail | 49 tok/s | 0.90 s | $0.500 | $3.60 | 262k | fp8 |
| DeepInfra | 44 tok/s | 0.61 s | $0.450 | $3.00 | 262k | fp8 |
| Phala | 43 tok/s | 2.08 s | $0.550 | $3.50 | 262k | – |
| Venice | 42 tok/s | 1.80 s | $0.750 | $4.50 | 128k | – |
| Alibaba Cloud Int. | 33 tok/s | 0.57 s | $0.390 | $2.34 | 262k | – |
| DigitalOcean | 10 tok/s | 1.35 s | $0.550 | $3.50 | 131k | – |
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 $0.220 per 1M. Reasoning tokens are not modelled.
| Workload | Tokens in / out | Cost | With caching | Time |
|---|---|---|---|---|
| Chat reply | 400 / 300 | $0.0004 | $0.0004 | 45.0 s |
| Summarise a 30-page report | 12,000 / 600 | $0.0027 | $0.0031 | 48.7 s |
| Code edit | 6,000 / 1,500 | $0.0026 | $0.0028 | 59.8 s |
| Agentic coding session | 60,000 / 4,000 | $0.014 | $0.017 | 1.5 min |
| Structured extraction | 2,000 / 200 | $0.0006 | $0.0006 | 43.8 s |
See also
Data as of 9 Sept 2026. Compare these configurations.