Alibaba Cloud (Qwen) · AI model
Qwen3.8-2.4T-A95B
Open-weights Qwen3.8 flagship: a sparse MoE with 2.4T total and about 95B activated parameters, text-only and thinking-only; hosted on Model Studio with a 1M-token context window.
Last verified Available 7 sources
Key specs
- Provider
- Alibaba Cloud (Qwen)
- Family
- Qwen3.8
- API identifier
- qwen3.8-2.4t-a95b
- Availability
- Available
- Release date
- 2026-08-12
- License
- Open weights (Qwen3.8-Max License)
- Context window
- 1,000,000 tokens
- Max output
- 131,072 tokens
- API access
- Yes
- Consumer app
- Unknown
Capabilities
Inputs
- Text input Yes
- Image input No Model card: text-only model; multimodal inputs are not supported.
- Audio input No
- Video input No
- File input Unknown
Outputs
- Text output Yes
- Image output No
- Audio output No
Features
- Reasoning Native Deep thinking page and model card: thinking mode only, cannot be disabled (the pricing table's 'Mode' column lists non-thinking and thinking; see open_questions).
- Tool calling Yes
- Web search Yes Hosted version on Model Studio.
"Unknown" means we have not verified this capability yet; it does not mean the model lacks it.
Pricing
Estimate your monthly cost →| Dimension | Tier | Price | Since | Verified | Source |
|---|---|---|---|---|---|
| Input tokens | base | $2.00 per 1M tokens | 2026-10-12 | 2026-10-25 | qwen3.8-2.4t-a95b Model Info - Alibaba Cloud Model Studio Singapore (International) list price for the hosted model; no input-length tiers (0 < tokens <= 1M). |
| Cached input tokens | base | $0.25 per 1M tokens | 2026-10-12 | 2026-10-25 | qwen3.8-2.4t-a95b Model Info - Alibaba Cloud Model Studio Input cache hit (implicit). |
| Cache writes | base | $2.50 per 1M tokens | 2026-10-12 | 2026-10-25 | qwen3.8-2.4t-a95b Model Info - Alibaba Cloud Model Studio Explicit cache creation. |
| Output tokens | base | $6.00 per 1M tokens | 2026-10-12 | 2026-10-25 | qwen3.8-2.4t-a95b Model Info - Alibaba Cloud Model Studio Output price (chain of thought + answer). |
| Tool and search requests | base | $10.00 per 1,000 requests | 2026-10-12 | 2026-10-25 | Web search with large models - Alibaba Cloud Model Studio Singapore: 'agent' search strategy, or the web_search tool via the Responses API or Anthropic-compatible method; per 1,000 calls, plus retrieved content billed as input tokens. No fee is listed for the default turbo/max strategies |
| Other | base | $0.17 per 1M units | 2026-10-12 | 2026-10-25 | qwen3.8-2.4t-a95b Model Info - Alibaba Cloud Model Studio Explicit cache hit. |
Prices are shown in the currency the provider publishes; we never convert currencies.
Benchmarks
| Benchmark | Setting | Score | Reported by | Measured | Verified | Source |
|---|---|---|---|---|---|---|
| GDPval-AA v2.1 Elo, higher is better | 'Qwen3.8 2.4T A95B'; effort not shown; Stirrup agent harness | 1,607 Elo | Third party | 2026-10-25 | 2026-10-25 | GDPval-AA v2.1 Leaderboard (Artificial Analysis) |
Elo is a relative rating: values change as models are added and the scale is re-anchored, so always store the date a rating was read. For example, Anthropic's launch page gives 1846 for Claude Opus 5.5, while the leaderboard on 2026-10-12 shows 1866 (Max, Default Fallback). Earlier versions (Google cites "GDPVal-AA v2") are not comparable. Judging is done by an LLM, not by human experts. Anthropic models are listed with "Default Fallback", meaning refused steps are routed to a fallback model. AA-Briefcase and GDP.pdf are separate evaluations.
One row per result, without rank numbers: each result was measured under its own setting (effort level, harness, tools, test version), as stated by the source, and scores compare only within the same setting. Self-reported results come from the provider and have not been reproduced independently.
Rankings
- #11 of 39 in Best AI for business as of 2026-10-26
Derived from our published ranking snapshots; each ranking explains its criteria and weights.
Compare Qwen3.8-2.4T-A95B
Guides
About Qwen3.8-2.4T-A95B
Qwen3.8-2.4T-A95B is the open-weights release of the Qwen3.8 flagship, using a sparse MoE architecture with 2.4 trillion total and about 95 billion activated parameters. It accepts text only and always reasons before answering; thinking cannot be disabled. The weights are published on Hugging Face under the custom Qwen3.8-Max License, which requires a separate license for Model-as-a-Service or AI work-assistant businesses whose revenue exceeds US$50 million over any 12-month period. Model Studio hosts it with a 1M-token context window, function calling, structured output and web search.
Strengths and weaknesses
Strengths
- Open weights of a 2.4T-parameter MoE model (about 95B activated)
- Hosted version: 1,000,000-token context window and up to 131,072 output tokens
- Function calling, structured output, web search and context caching on Model Studio
Weaknesses
- Text-only input (no image or video), unlike the hosted qwen3.8-max
- Thinking cannot be disabled; maximum chain-of-thought length is 131,072 tokens versus 262,144 for qwen3.8-max
- Custom license with additional conditions for large commercial users
Sources
- qwen3.8-2.4t-a95b Model Info - Alibaba Cloud Model Studio
- Model lifecycle and updates - Alibaba Cloud Model Studio
- Qwen3.8-Max License (Qwen3.8-2.4T-A95B LICENSE)
- Qwen/Qwen3.8-2.4T-A95B - Hugging Face model card
- Use deep thinking models via API | Model Studio - Alibaba Cloud Model Studio
- Web search with large models - Alibaba Cloud Model Studio
- GDPval-AA v2.1 Leaderboard (Artificial Analysis)
Last verified Page updated 2026-10-26