Qwen released Qwen3.8-Flash-Next, the first open-weight model built on the architecture that will underpin Qwen4. It's a roughly 125B-parameter mixture-of-experts with only ~6B active per token, targeting extreme cost-efficiency while reading images natively with a 262,144-token context window. Official FP8 weights are out, and the model posts agentic-coding scores in Claude Opus territory at a fraction of the training cost.

Qwen · 2026-08-26 · Read the dispatch →