Alibaba to open-source medium Qwen3.6 models

Qwen is doubling down on its "proprietary-first, open-soon" strategy, keeping pressure on the Llama and Mistral ecosystems. Medium-sized models likely in the 32B-70B range provide a sweet spot for local deployment and enterprise fine-tuning. Native Chain-of-Thought reasoning and a 1M context window in open-weights format will lower the barrier for building complex autonomous agents, while a 78.8 SWE-bench Verified score signals serious contention for top coding benchmarks.