Qwen/Qwen3.8-2.4T-A95B-FP8
Parameters2446B (95B active)
Context256K
Input → outputtext → text
Licenseother
PrecisionF8_E4M3 · fp8
ArchitectureQwen3_5MoeForCausalLM
ReleasedAug 8, 2026
Downloads8.1K
Based on Qwen/Qwen3.8-2.4T-A95B
Other sizes and formats in the family
| Model | Params | Precision | Downloads |
|---|---|---|---|
| Qwen/Qwen3.8-2.4T-A95B | 2446B | BF16 | 32.5K |
| Qwen/Qwen3.8-27B | 27.8B | BF16 | 6.8M |
| Qwen/Qwen3.8-27B-FP8fp8 | 27.8B | F8_E4M3 | 4.5M |
This release in the news
No coverage of this exact release yet.
Recent news about Qwen
- Oct 11, 2026 · Opinion / analysisEvery Model That Can Be Run On 10-16GB VRAM Ranked
- Oct 11, 2026 · Opinion / analysisReverse Engineering w/ Local?
- Oct 11, 2026 · Benchmark resultAgents That Own Their Inference — Du'an Lightfoot & Khaja Omer, Akamai Technologies
- Oct 11, 2026 · Opinion / analysisUPDATE: Qwen 3.8 27B 140 tok/s on single RTX 3090 Megakernel: KL divergence 0.0009 vs llama.cpp
- Oct 11, 2026 · Model releaseQwen/Qwen-Image-2.1-Turbo
- Oct 11, 2026 · Opinion / analysisBuilding prompts with LLMs
- Oct 11, 2026 · Opinion / analysis[P] Pecision models that score every allowed label from the logits: Jebadiah v2.1 (27B, 9B), open weights and self-run benchmark results [P]
- Oct 11, 2026 · Opinion / analysisPSA: DeepSeek V4.1 Flash habitually exfiltrates API keys. It is dangerously misaligned and may be hazardous to use