Modelofficial release

Qwen/Qwen3.8-27B-FP8

Qwen →Hugging Face ↗
Parameters27.8B
Context256K
Input → outputimage + text → text
Licenseapache-2.0
PrecisionF8_E4M3 · fp8
ArchitectureQwen3_5ForConditionalGeneration
ReleasedAug 13, 2026
Downloads4.5M

Based on Qwen/Qwen3.8-27B

Other sizes and formats in the family

ModelParamsPrecisionDownloads
Qwen/Qwen3.8-2.4T-A95B2446BBF1632.5K
Qwen/Qwen3.8-2.4T-A95B-FP8fp82446BF8_E4M38.1K
Qwen/Qwen3.8-27B27.8BBF166.8M

Compare side by side →

This release in the news

No coverage of this exact release yet.

Recent news about Qwen

  1. Oct 11, 2026 · Opinion / analysis
    Every Model That Can Be Run On 10-16GB VRAM Ranked
  2. Oct 11, 2026 · Opinion / analysis
    Reverse Engineering w/ Local?
  3. Oct 11, 2026 · Benchmark result
    Agents That Own Their Inference — Du'an Lightfoot & Khaja Omer, Akamai Technologies
  4. Oct 11, 2026 · Opinion / analysis
    UPDATE: Qwen 3.8 27B 140 tok/s on single RTX 3090 Megakernel: KL divergence 0.0009 vs llama.cpp
  5. Oct 11, 2026 · Model release
    Qwen/Qwen-Image-2.1-Turbo
  6. Oct 11, 2026 · Opinion / analysis
    Building prompts with LLMs
  7. 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]
  8. Oct 11, 2026 · Opinion / analysis
    PSA: DeepSeek V4.1 Flash habitually exfiltrates API keys. It is dangerously misaligned and may be hazardous to use