Qwen3.8-2.4T-A95B
Alibaba·
Overview
Open-weight checkpoint of the Max-class flagship and the first Qwen-Max-class model with downloadable weights: 2.4T total / 95B active sparse MoE, published on Hugging Face and ModelScope alongside an FP8 build and served via API. It is not the hosted Qwen3.8-Max — the checkpoint is text-only (no vision or video input) and always reasons, thinking mode cannot be disabled. Native context is 262,144 tokens, extensible to roughly 1M via YaRN. Shipped under the bespoke Qwen3.8-Max License rather than Apache 2.0. In FP8 the weights occupy about 2,325 GiB, so self-hosting takes a 16-GPU class deployment.
Capabilities and innovations
Benchmarks
| Benchmark | Score | Source |
|---|---|---|
| GPQA Diamond | 93.54% | third party GPQA Diamond 93.54%, measured by Artificial Analysis against the open-weight checkpoint itself (own record: open weights, 2.4T/95B, Qwen3.8-Max License), not against the hosted Qwen3.8-Max API which it measures separately at 92.73. |
| Humanity’s Last Exam (no tools) | 42.45% | third party HLE 42.45% no-tools, Artificial Analysis, same checkpoint record (hosted Qwen3.8-Max: 43.05). |
| Terminal-Bench 2.x | 82.02% | third party Terminal-Bench 2.1 82.02%, Artificial Analysis, same checkpoint record (hosted Qwen3.8-Max: 81.27). Same source and harness as the DeepSeek-V4 and Qwen3.8-Max entries. |
- SWE-bench Pro: independent evaluation pending — Alibaba published no benchmark table for the open-weight checkpoint, and neither Vals AI nor Artificial Analysis runs SWE-bench Pro.
- MMLU-Pro: independent evaluation pending — Vals AI has no entry for this checkpoint as of 2026-08-17 (it does cover the hosted Qwen3.8-Max at 88.60).
- MMMU-Pro: not reported by the vendor — Not applicable: the open-weight checkpoint has no vision path, unlike the hosted Qwen3.8-Max API. Artificial Analysis reports mmmuPro as null for it, which is consistent.
- Terminal-Bench 3.0: not reported by the vendor
API pricing
$2 per million input tokens, $6 per million output tokens (USD, cheapest OpenRouter endpoint, checked 2026-08-17)
Architecture and hardware
- Parameters
- 2400B total, 95B active per token (MoE)
- Estimated VRAM at Q4
- ~1380 GB, Frontier class
Links
More from Alibaba
Data curated by AI Model Timeline. See the methodology for admission criteria, benchmark eras and source priority.