AI;DR
AI;DR
← Back to live feed1 story

Monday, Aug 17, 2026

1

Alibaba Qwen3.8 27B Matches GPT 5.6 Luna Performance in First Local Model Frontier Score

The latest Artificial Analysis Intelligence Index shows a score of 52 for the new open weight release from Alibaba, placing it alongside industry leading models like DeepSeek V4-Pro. This benchmark puts the 27B parameter tool in a performance tier previously reserved for proprietary non local systems.

The model is natively multimodal and licensed under Apache 2.0 to enable broad developer access. Users running the model on a single Nvidia RTX 5090 with 32 GB VRAM reported speeds of 115 tokens per second, while optimized recipes from SGLang support 206.1 tokens per second.

Sign in to suggest edits

Key sources

  1. SOURCE@cline“We weren’t expecting this pace of local progress anywhere near this soon”x.com
  2. SUPPORT@xenovacom“can run locally... even in your browser with custom WebGPU kernels!”x.com
  3. SUPPORT@sgl_project“206.1 tok/s decode on a single RTX 5090, with our NVFP4 plus DSpark”x.com
  4. SUPPORT@rohanpaul_ai“Qwen3.8-27B's official BF16 checkpoint is 55.6 GB”x.com
Markdown