qwen
Qwen: Qwen3.8 2.4T A95B
qwen/qwen3.8-2.4t-a95b
Our take
AI-assisted analysis based on the sources below. Suitability is an editorial assessment; test it on your own workload.
Qwen3.8 2.4T A95B (qwen/qwen3.8-2.4t-a95b) is Qwen's open-weight sparse mixture-of-experts model, activating 95 billion of 2.4 trillion total parameters, and is described as the open-weight counterpart of Qwen3.8 Max. It is text-only, combines gated DeltaNet and gated attention layers with 512 experts, and offers a native 262,144-token context documented as extensible to roughly one million tokens. Reasoning is always enabled with adjustable effort, and tool calling plus structured outputs are supported. Independent measurements are limited to catalog-reported indices.[1][2]
Strengths
- Very large context window: 262,144 tokens natively, documented as extensible to about one million tokens, suiting long-document and repository-scale inputs.[1][2]
- Reasoning effort is tunable across three levels and prior reasoning context can be preserved across turns; tool calling and structured outputs are supported.[1][2]
- Distributed in Hugging Face Transformers format with stated compatibility for vLLM, SGLang and TokenSpeed, easing self-hosted deployment.[2]
Limitations
- Text-only input and output; vision input and an optional non-thinking mode are listed as features of the separate Qwen3.8-Max, while reasoning is always on for this variant.[1][2]
- Published benchmark tables cover Qwen3.8-Max, not this variant; independent evidence here is limited to catalog coding and agentic indices with no intelligence index, so validate on your own workloads.[1][2]
- Knowledge cutoff is not reported, and the model uses a custom qwen3.8-max license whose terms are not detailed in this source pack.[1][2]
Where to use it
- Repository-scale coding and refactoring agents that can benefit from the long context and tool support; a suggested fit, not benchmark-verified for this variant.[1][2]
- Long-horizon, multi-step agentic workflows where adjustable reasoning effort and preserved thinking context may help reliability.[1][2]
- Self-hosted deployments seeking an open-weight, Max-class Qwen model via vLLM or SGLang, subject to validating quality and infrastructure fit.[2]
Sources
- [1] OpenRouter catalog · Qwen: Qwen3.8 2.4T A95BSources checked: Nov 13, 2026
- [2] Model documentation · Qwen: Qwen3.8 2.4T A95BSources checked: Nov 13, 2026
About this model
Description below is the original text supplied by OpenRouter, shown in the source language.
Qwen3.8 2.4T A95B is an open-weight sparse mixture-of-experts model from Qwen and the open-weight variant of [Qwen3.8 Max](/qwen/qwen3.8-max), with 95 billion active parameters out of 2.4 trillion total. It is...
Specifications
- Model ID
- qwen/qwen3.8-2.4t-a95b
- Canonical ID
- qwen/qwen3.8-2.4t-a95b-20260812
- Creator
- qwen
- Added to OpenRouter
- 12.08.2026
- Context length
- 1,048,576
- Modality
- text->text
- Input modalities
- text
- Output modalities
- text
- Token pricing
- USD per 1M tokens: in $2 / out $6
- Knowledge cutoff
- —
- Expiration date
- —
Tasks and capabilities
A grade appears only where the source supplies a matching measurement for that task, the model is active and the cohort holds at least 10 comparable models. The capability column lists facts the source declares — supported formats, reasoning and tool parameters — which are not measured quality.
| Task | Grade | Source-declared capability |
|---|---|---|
| General quality | No measurement | reasoning: mandatory · default xhigh · efforts: xhigh, medium, low · parameters: include_reasoning, reasoning, reasoning_effort |
| Coding | AA — next 20% of the cohort | — |
| Agents | SS — top 10% of the cohort | parameters: structured_outputs, tool_choice, tools |
| Writing | No measurement | output: text |
| Reasoning | No measurement | reasoning: mandatory · default xhigh · efforts: xhigh, medium, low · parameters: include_reasoning, reasoning, reasoning_effort |
| Document analysis | No measurement | — |
| Vision | No measurement | — |
| Image generation | No measurement | — |
| Audio | No measurement | — |
| Video | No measurement | — |
| Embeddings | No measurement | — |
Evaluations
| Source | Benchmark | Value | Grade | Data retrieved |
|---|---|---|---|---|
| Artificial Analysis | Coding index | 71.9 | AA — next 20% of the cohort | 13.11.2026 |
| Artificial Analysis | Agentic index | 50.1 | SS — top 10% of the cohort | 13.11.2026 |
How grades work
Each letter grade compares a model only against other models scored by the same source and metric in the current snapshot. S means the top 10%, A the next 20%, B the next 30%, C the rest. Models without a published measurement are shown as not rated — a supported format alone never implies measured quality.
- S — S — top 10% of the cohort
- A — A — next 20% of the cohort
- B — B — next 30% of the cohort
- C — C — below the B band
Providers
Optional live provider statistics from OpenRouter. Missing values are unknown, not zero.
| Provider | Context length | Price in /1M | Price out /1M | Latency p50 (30m, s) | Throughput p50 (30m, tok/s) | Uptime (30m) |
|---|---|---|---|---|---|---|
| Novita | 1,000,000 | $2 | $6 | — | — | 100% |
| Alibaba | 1,000,000 | $2 | $6 | — | — | 100% |
| SiliconFlowfp8 | 1,048,576 | $2 | $6 | — | — | 100% |
| Venice | 262,144 | $2 | $6 | — | — | 0% |
Provider data retrieved: 13.11.2026
Related articles
No published articles identify this model yet.