New Qwen 3.6 is INSANE FREE!
Qwen 3.6 Plus is a new free AI model from Alibaba that's currently available during a preview period with no cost. The model features a 1 million token context window, built-in chain of thought reasoning, and strong performance in coding and agent workflows, competing with top-tier AI models while being completely free to use.
Summary
Julian Goldie introduces Qwen 3.6 Plus, a new AI model from Alibaba that he claims is being overlooked despite its powerful capabilities. Qwen is Alibaba's AI model family that has been consistently improving with each release, similar to DeepSeek in disrupting the AI space. Unlike previous open-source Qwen models, 3.6 Plus is currently a closed preview without public weights available. The model represents a significant upgrade from the 3.5 Plus series released in March 2026, built on a hybrid architecture for improved efficiency and capability. Key features include a massive 1 million token context window for handling entire codebases and long documents, up to 65,000 output tokens for complete responses, always-on chain of thought reasoning, and stable tool use and function calling capabilities. The hybrid architecture combines linear attention with sparse mixture of experts, activating only necessary parts for specific tasks. The model excels particularly in agentic coding and front-end development. Currently free on Open Router during the preview period, users pay zero cost for both input and output tokens, though prompts may be used for model improvement. Goldie emphasizes that the quality gap between free and premium AI models is shrinking rapidly. Practical applications include code review, front-end generation, long document analysis, AI agent building, and complex reasoning tasks. He recommends testing the model with real workflows and providing maximum context for best results, while promoting his AI communities for deeper learning.
Key Insights
- Qwen 3.6 Plus fixes the major complaint about the 3.5 series where it sometimes overthought problems and took too long, now reaching conclusions faster and using fewer tokens
- The quality difference that used to justify paying premium prices for proprietary models is shrinking fast, meaning you're not making a compromise by using a free model anymore
- Early testers have specifically called out how stable the tool calling is compared to previous Qwen versions, with less flakiness meaning fewer retries when building real workflows
- The hybrid architecture combines linear attention and sparse mixture of experts, only activating the parts of the model it needs for a given task, which is how it handles massive inputs without grinding to a halt
- Qwen 3.6 Plus performs at or above leading state-of-the-art models according to Alibaba's own benchmark results, with early independent analysis showing it edging ahead of Qwen 3.5 Plus in overall capability and consistency
Topics
Full transcript available for MurmurCast members
Sign Up to Access