AlibabaQwen3.5 122B-A10B
Best configuration ranks #166 of 610 on the BenchLeader Index at 56.0 ±5.1. Last measured 8 Sept 2026. Released 24 Feb 2026.
- Blended price
- $1.10/M
- $0.400 in · $3.20 out
- Output speed
- 129 tok/s
- First answer
- 18 s
- first token 0.70 s
- Context
- 262k
- Overall index56
- Reasoning60
- Coding48
- Agents & tools61
- Knowledge44
- Instruction following73
- Human preference60
- Multimodal59
- Long context65
- Composite49
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 | 50.0 | #297 | 142 tok/s | 2.32 s | $0.0011 | Agents & tools 59 · Coding 38 · Composite 51 · Instruction following 52 · Knowledge 38 · Long context 57 · Multimodal 55 |
| defaultbest | 56.0 | #166 | 129 tok/s | 18 s | $0.0011 | Agents & tools 61 · Coding 48 · Composite 49 · Human preference 60 · Instruction following 73 · Knowledge 44 · Long context 65 · Multimodal 59 · Reasoning 60 |
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 | Trend |
|---|---|---|---|---|
| LMArena Hard Prompts | – | 1433#113 | LMArena | |
| GPQA Diamond (AA)not in index | 82.7%#165 | 85.7%#119 | Artificial Analysis | |
| Humanity's Last Exam (AA)not in index | 15.9%#186 | 25.2%#134 | Artificial Analysis |
Coding
| Benchmark | no reasoning | default | Source | Trend |
|---|---|---|---|---|
| SciCode | 35.6%#131 | – | SciCode | |
| LMArena Coding | – | 1459#115 | LMArena | |
| LMArena WebDev | – | 1358#87 | LMArena | |
| SciCode (AA)not in index | – | 39.7%#117 | Artificial Analysis |
Agents & tools
| Benchmark | no reasoning | default | Source | Trend |
|---|---|---|---|---|
| Terminal-Bench Hard | 29.6%#114 | 31.1%#104 | Artificial Analysis | |
| τ²-Bench Telecom (AA)not in index | 84.5%#99 | 93.6%#42 | Artificial Analysis |
Knowledge
| Benchmark | no reasoning | default | Source | Trend |
|---|---|---|---|---|
| AA-Omniscience | -55.0#360 | -41.5#259 | Artificial Analysis |
Instruction following
| Benchmark | no reasoning | default | Source | Trend |
|---|---|---|---|---|
| IFBench | 50.8%#167 | 75.7%#27 | Artificial Analysis |
Human preference
| Benchmark | no reasoning | default | Source | Trend |
|---|---|---|---|---|
| LMArena Text | – | 1417#110 | LMArena |
Multimodal
| Benchmark | no reasoning | default | Source | Trend |
|---|---|---|---|---|
| LMArena Vision | – | 1245#54 | LMArena | |
| MMMU-Pro | 70.3%#114 | 75.0%#78 | Artificial Analysis |
Long context
| Benchmark | no reasoning | default | Source | Trend |
|---|---|---|---|---|
| AA-LCR | 61.3%#210 | 76.3%#100 | Artificial Analysis |
Composite
| Benchmark | no reasoning | default | Source | Trend |
|---|---|---|---|---|
| AA Intelligence Index | 17.8#196 | 16.2#211 | 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. | 115 tok/s | 0.53 s | $0.260 | $2.08 | 262k | – |
| DeepInfra | 106 tok/s | 0.45 s | $0.290 | $2.40 | 262k | fp4 |
| NovitaAI | 91 tok/s | 0.70 s | $0.400 | $3.20 | 262k | bf16 |
| AtlasCloud | 89 tok/s | 1.10 s | $0.300 | $2.40 | 262k | fp8 |
| SiliconFlow | 78 tok/s | 0.99 s | $0.260 | $2.08 | 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 | $0.0011 | – | 20.1 s |
| Summarise a 30-page report | 12,000 / 600 | $0.0067 | – | 22.5 s |
| Code edit | 6,000 / 1,500 | $0.0072 | – | 29.4 s |
| Agentic coding session | 60,000 / 4,000 | $0.037 | – | 48.8 s |
| Structured extraction | 2,000 / 200 | $0.0014 | – | 19.4 s |
See also
Data as of 9 Sept 2026. Compare these configurations.