Alibaba's Qwen team has open-sourced Qwen3.8-27B, a 27-billion parameter model that has achieved top rankings on several benchmarks, including SWE Bench Pro and OSWorld. This model demonstrates significant advancements in agentic capabilities, such as planning and multi-step task completion, outperforming larger models and even cloud-based flagships. Its architecture incorporates a novel blend of linear and full attention layers to efficiently handle long contexts, and it offers controllable "thinking" modes for users. The model's multimodal training allows it to operate effectively across various interfaces like desktops, browsers, and mobile devices. AI
IMPACT Sets new SOTA on agentic benchmarks, demonstrating powerful local model capabilities and pushing cloud model performance.
RANK_REASON Frontier-lab model release with system card and benchmark results. [lever_c_demoted from frontier_release: ic=2 ai=1.0]
- Qwen3.8-27B
- Agents Last Exam
- Alibaba Group
- AndroidWorld
- Anthropic
- CoWorkBench
- LiveCodeBench v6
- Opus4.6 Max
- OSWorld
- Qwen
- RecreationBench
- SWE Bench Pro
- WebArena
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
