Qwen3.8: A 2.4T-Parameter Open Model That Beats GPT-5.6 Sol on PaperBench

Qwen/Qwen3.8-2.4T-A95B

Qwen releases Qwen3.8-2.4T-A95B, its most capable open model yet, with 2.4 trillion total parameters and 95 billion activated. Built on Qwen3.5's architecture, it excels in coding, professional work, research, and long-horizon agentic tasks. Benchmark results show it outperforms GPT-5.6 Sol on PaperBench (93.0 vs 90.5) and rivals Claude Opus 4.8 on several coding benchmarks. The model supports flexible reasoning control and is compatible with vLLM, SGLang, and TokenSpeed.