China AI Hub
Models / Qwen3.8-2.4T-A95B
Qwen3.8-2.4T-A95B
Qwen3.8 family · Provider: alibaba-cloud · Status: active · Released: 2026-08-12
Qwen3.8-2.4T-A95B is a Qwen3.8 model developed by alibaba-cloud, a Chinese AI company, released 2026-08-12 with a 262,144-token context window and open weights.
| Key fact | Value |
|---|---|
| Model ID | qwen3.8-2.4t-a95b |
| Architecture | 2.4T-parameter MoE, 95B activated, 512 experts (10 routed + 1 shared per token), 92 layers, Gated DeltaNet + Gated Attention hybrid |
| Context window | 262,144 tokens |
| Open weights | Yes |
| License | Qwen3.8-Max License (custom MIT-style: unrestricted use/copy/modify/sell, but products with >100M MAU or >US$20M/month revenue must display the model name; Model-as-a-Service or AI Work Assistant businesses with >US$50M/12-month revenue need a separate license from Qwen) |
| Self-hosting | Yes |
| API available | No |
Capabilities
| Capability | Supported |
|---|---|
| Reasoning | Yes |
| Coding | Unknown |
| Math | Unknown |
| Chinese | Unknown |
| English | Unknown |
| Multilingual | Unknown |
| Vision | No |
| Audio | Unknown |
| Video | Unknown |
| Tool calling | Unknown |
| Function calling | Unknown |
| Structured output | Unknown |
| Agent capability | Unknown |
| RAG | Unknown |
| Computer use | Unknown |
Known limitations
- Text-only input; thinking cannot be disabled; reasoning_effort xhigh/medium/low
- Native context 262,144 tokens, extensible to 1,010,000
- Not the same product as the qwen3.8-max API model (which adds vision/video input, non-thinking mode and 1M default context)
Qwen3.8-2.4T-A95B is Qwen’s first Qwen-Max-class open-weight release (2026-08-12): a 2.4T-parameter MoE with 95B activated parameters using a Gated DeltaNet + Gated Attention hybrid. It is text-only and thinking-only (reasoning cannot be disabled; reasoning_effort xhigh/medium/low), with a native 262,144 context extensible to 1,010,000 tokens.
The license is a custom MIT-style agreement with commercial attribution and revenue-threshold requirements. The hosted qwen3.8-max API model is the related closed product with broader modalities.
Provider
Related models (same family)
API
Model family timeline
| Model | Released | Status |
|---|---|---|
| Qwen3.8-Max | 2026-08 | active |
| Qwen3.8-Flash | 2026-08 | active |
| Qwen3.8-2.4T-A95B (this page) | 2026-08-12 | active |
Data interpretation
| Field | Value | Evidence type |
|---|---|---|
| Context window | 262,144 tokens | Official |
| Architecture | 2.4T-parameter MoE, 95B activated, 512 experts (10 routed + 1 shared per token), 92 layers, Gated DeltaNet + Gated Attention hybrid | Official |
| Open weights | Yes | Official |
| License | Qwen3.8-Max License (custom MIT-style: unrestricted use/copy/modify/sell, but products with >100M MAU or >US$20M/month revenue must display the model name; Model-as-a-Service or AI Work Assistant businesses with >US$50M/12-month revenue need a separate license from Qwen) | Official |
| Family position | 3 models in the Qwen3.8 family | China AI Hub analysis |
Evidence types: Official = vendor documentation, pricing pages or model cards. Vendor-reported = benchmark scores published by the vendor. China AI Hub analysis = derived from the database itself. See the sourcing policy.
Sources
Confidence and source hierarchy per the sourcing policy. Facts change; check the source before relying on this page.
What is the context window of Qwen3.8-2.4T-A95B?
Qwen3.8-2.4T-A95B has a 262,144-token context window.
Is Qwen3.8-2.4T-A95B open weight?
Yes — Qwen3.8-2.4T-A95B weights are openly available under the Qwen3.8-Max License (custom MIT-style: unrestricted use/copy/modify/sell, but products with >100M MAU or >US$20M/month revenue must display the model name; Model-as-a-Service or AI Work Assistant businesses with >US$50M/12-month revenue need a separate license from Qwen) license.
How much does Qwen3.8-2.4T-A95B cost through the API?
API pricing for Qwen3.8-2.4T-A95B is not currently listed in our database.
Where does China AI Hub get its Qwen3.8-2.4T-A95B data?
From 3 sources (official pages first), last verified 2026-09-20. Benchmark scores are labeled by source type; see the sourcing policy for details.