Recently, the open-source field has made significant progress. The Qwen3.8-2.4T-A95B model weights have been officially released by the Qwen large model team at Alibaba. This is also the first time that the team has fully opened its flagship Max-level model to developers and researchers around the world.
As the core of this open source, the model has a total parameter count of 2.4 trillion, with 95 billion parameters activated per token. It natively supports an ultra-long context of 260,000 tokens and can be further expanded to 1,010,000 tokens. Its underlying architecture continues the advantages of Qwen3.5, achieving comprehensive performance improvements in programming, daily office work, research assistance, and long-term intelligent agent tasks.
In various authoritative benchmark tests, the model has shown outstanding performance. Its results rank among the top models in the PaperBench paper reproduction benchmark and the OSWorld-Verified computer operation capability evaluation. In some indicators, it even surpasses several international mainstream models. In addition, its strong long-term autonomous working capabilities are remarkable, as it can independently complete complex programming tasks lasting several days and autonomously reproduce and optimize research papers.
