AlibabaQwen3.8 2.4T A95B
Qwen3.8 2.4T A95B is an Alibaba open-weights reasoning model, released 12 Aug 2026. Its best configuration ranks #63 of 756 on the BenchLeader Index at 63.5 ±5.1, in the upper half. The ± is the point: 153 other configurations score within that range, so they and this one cannot be told apart on quality alone — price and speed are what separate them. It scores highest in agents & tools (79) and lowest in knowledge (64). At $3.00 per million tokens blended it is pricier than most ranked models. Output speed of 40 tokens per second puts it in the slowest quarter, with a first answer in 53.3 s. It has been measured at 2 reasoning-effort settings; this summary describes the best-scoring one, and the tabs above switch between them.
- Blended price
- $3.00/M
- $2.00 in · $6.00 out
- Output speed
- 40 tok/s
- measured by Artificial Analysis
- First answer
- 53 s
- first token 1.48 s
- Context
- 262k
- Full answer
- 66 s
- median, reasoning included
- Cost per run
- $2.16
- one full Intelligence Index run
- Released
- 12 Aug 2026
- Overall index64
- Reasoning75
- Agents & tools79
- Knowledge64
- Long context66
- Composite75
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. “Not stated” collects results from publishers that did not say which setting they used; for a reasoning model that is usually its thinking mode, but we do not assume it. Pick a setting here or at the top of the page to see its price, speed and category scores.
| Effort | Index | Rank | Speed | First answer | Chat reply cost | Categories |
|---|---|---|---|---|---|---|
| xhigh | – | – | 40 tok/s | 53 s | $0.0026 | Agents & tools 65 |
| not statedbest | 63.5 | #63 | 40 tok/s | 53 s | $0.0026 | Agents & tools 79 · Composite 75 · Knowledge 64 · Long context 66 · Reasoning 75 |
How its index has moved
19 Sept 2026 to 2 Oct 2026The index is recomputed from scratch every day, so a line moves when a new benchmark result lands, when a publisher revises a score, or when the models it is normalised against change. Early movement usually means the score is still settling.
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 | xhigh | not stated | Source | Trend |
|---|---|---|---|---|
| GPQA Diamond (AA)not in index | – | 93.5%#14 | Artificial Analysis | |
| Humanity's Last Exam (AA)not in index | – | 42.5%#60 | Artificial Analysis | |
| CritPt | – | 20.0%#56 | Artificial Analysis |
Coding
| Benchmark | xhigh | not stated | Source | Trend |
|---|---|---|---|---|
| SciCode (AA)not in index | – | 54.0%#62 | Artificial Analysis | |
| Terminal-Bench 4.0 (AA)not in index | – | 11.1%#74 | Artificial Analysis | |
| Terminal-Bench 2.1 (AA)not in index | – | 82.0%#39 | Artificial Analysis |
Agents & tools
| Benchmark | xhigh | not stated | Source | Trend |
|---|---|---|---|---|
| MCP Atlas | 84.5%#4 | – | Scale AI SEAL | |
| GDPval (AA)not in index | – | 54.9%#26 | Artificial Analysis | |
| τ²-Bench Banking (AA) | – | 49.1%#5 | Artificial Analysis |
Knowledge
| Benchmark | xhigh | not stated | Source | Trend |
|---|---|---|---|---|
| AA-Omniscience | – | 4.3#103 | Artificial Analysis | |
| AA-Omniscience: accuracynot in index | – | 31.3%#159 | Artificial Analysis | |
| AA-Omniscience: non-hallucinationnot in index | – | 60.8%#66 | Artificial Analysis |
Long context
| Benchmark | xhigh | not stated | Source | Trend |
|---|---|---|---|---|
| AA-LCR | – | 80.3%#63 | Artificial Analysis |
Composite
| Benchmark | xhigh | not stated | Source | Trend |
|---|---|---|---|---|
| AA Intelligence Index | – | 39.9#55 | Artificial Analysis |
Safety & honesty
| Benchmark | xhigh | not stated | Source | Trend |
|---|---|---|---|---|
| FORTRESSnot in index | 21.6%#34 | – | Scale AI SEAL |
Where it wins
Benchmarks where this configuration ranks in the top five of every configuration measured.
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 |
|---|---|---|---|---|---|---|
| Modal | 206 tok/s | 0.46 s | $2.00 | $6.00 | 1M | – |
| Together | 162 tok/s | 2.96 s | $2.00 | $6.00 | 1.0M | – |
| DeepInfra | 82 tok/s | 0.93 s | $2.00 | $6.00 | 262k | fp4 |
| Venice | 51 tok/s | 1.14 s | $2.00 | $6.00 | 262k | – |
| Alibaba Cloud Int. | 42 tok/s | 1.48 s | $2.00 | $6.00 | 1M | – |
| NovitaAI | 33 tok/s | 4.26 s | $2.00 | $6.00 | 1M | – |
| SiliconFlow | 32 tok/s | 7.10 s | $2.00 | $6.00 | 1.0M | 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 $0.250 per 1M. Reasoning tokens are not modelled.
| Workload | Tokens in / out | Cost | With caching | Time |
|---|---|---|---|---|
| Chat reply | 400 / 300 | $0.0026 | $0.0021 | 1.0 min |
| Summarise a 30-page report | 12,000 / 600 | $0.028 | $0.012 | 1.1 min |
| Code edit | 6,000 / 1,500 | $0.021 | $0.013 | 1.5 min |
| Agentic coding session | 60,000 / 4,000 | $0.144 | $0.065 | 2.6 min |
| Structured extraction | 2,000 / 200 | $0.0052 | $0.0026 | 58.3 s |
See also
Data as of 5 Oct 2026. Compare these configurations.
Cite as: BenchLeader, “Qwen3.8 2.4T A95B: benchmarks, pricing, speed and rank”, https://www.benchleader.com/models/qwen3-8-2-4t-a95b, data as of 5 Oct 2026.