QWEN SEA

Alibaba's Qwen family — the most widely fine-tuned open weights in agent stacks. Current lineup on the chart: Qwen3.8-Flash-Next (the August 2026 open-weights release previewing the Qwen4 architecture — a 125B-parameter hybrid MoE with 6B parameters activated per token), Qwen3 (the latest general-purpose base) and Qwen3-Coder (a code-specialized variant tuned for repository-scale agent work). Released under permissive licenses that make it the default base for downstream coding agents and tool-use fine-tunes.