Qwen3.8-27B open-source model rivals Claude Opus on benchmarks

Qwen has released its Qwen3.8-27B model, an open-source model that rivals the performance of larger, proprietary models like Claude Opus 4.6 Max on various benchmarks, particularly in coding and agent tasks. This 27-billion parameter model is designed to run on consumer-grade GPUs, offering a 262K native context window that can be extended to 1 million tokens. It also features native multimodal capabilities, allowing it to process images, PDFs, and even video, with customizable reasoning depths for efficiency. AI

IMPACT Accelerates the availability of high-performance, multimodal models for local deployment, potentially lowering the barrier for complex AI agent development.

RANK_REASON Open-source release of a new frontier model by a major lab (Qwen) with benchmark results competitive with top proprietary models. [lever_c_demoted from frontier_release: ic=1 ai=1.0]

Read on 量子位 (QbitAI) →

• Claude
• Claude Opus 4.6 Max
• Hugging Face
• NVIDIA GH200
• Qwen
• Qwen3.8-27B
• QwenSWEBench
• SGLang
• SWE-bench Pro
• TokenSpeed
• Transformers
• vLLM

• model release
• product

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →