Model release
Qwen3.8 Flash activates just 6B parameters
Qwen’s comparison shows the Flash model beating Qwen3.8 27B across every listed benchmark and Qwen3.7 Plus on eight of fourteen.
Qwen3.8 Flash activates only 6B parameters per token, compared with 27B for Qwen3.8 27B. In Qwen’s published comparison, the Flash model leads the 27B model on every benchmark shown.
It also beats Qwen3.7 Plus on eight of fourteen benchmarks while activating 6B parameters rather than 17B. That is the standout efficiency claim in Qwen’s table.
The model uses the architecture Qwen says it is developing towards Qwen 4.