QWQ-Max
<think>...</think> with Qwen
386 followers
<think>...</think> with Qwen
386 followers
QwQ-Max-Preview from Qwen is a powerful new LLM excelling in reasoning, math, coding, and agent tasks. Features a "thinking mode" for complex problems. Open-source coming soon!
This is the 5th launch from QWQ-Max. View more
Qwen3.6-Max-Preview
Launching today
Qwen3.6-Max-Preview is Alibaba's next frontier model — stronger agentic coding, world knowledge, and instruction following than Qwen3.6-Plus. #1 on six major coding benchmarks. Available now on Qwen Studio and via API. Still under active development.
Free
Launch Team
Qwen3.6-Max-Preview is Alibaba's sharpest model yet — and it's still getting better.
Most frontier models improve one thing at a time. Qwen3.6-Max-Preview improves agentic coding, world knowledge, and instruction following all at once — and ships as a preview, meaning further gains are already on the way.
What makes it different:
It tops six major coding benchmarks simultaneously — SWE-bench Pro, Terminal-Bench 2.0, SkillsBench, QwenClawBench, QwenWebBench, and SciCode — while also closing gaps in knowledge and tool-calling reliability that most coding-focused models ignore.
Key features:
#1 on six major coding benchmarks — biggest gains in SkillsBench (+9.9) and SciCode (+6.3)
Stronger world knowledge — SuperGPQA +2.3, QwenChineseBench +5.3 over Qwen3.6-Plus
Better instruction following — ToolcallFormatIFBench +2.8 for more reliable agent execution
preserve_thinking support — maintains full reasoning context across turns for agentic tasks
256K context window — handles long codebases and documents without losing context
OpenAI and Anthropic API compatible — drops into existing workflows without changes
$1.3/1M input, $7.8/1M output — frontier capability at accessible pricing
Who it's for & use cases:
Developers building coding agents and front-end tooling
Teams running knowledge-intensive AI applications
Anyone already on Qwen3.6-Plus looking for a meaningful upgrade
Why it matters:
This is a preview — not a finished model. The benchmarks already beat the current generation, and Alibaba is still iterating. Getting in early means shaping what comes next.
Available now on Qwen Studio. API via Alibaba Cloud Model Studio coming soon.
P.S. I hunt the latest and greatest launches in AI, SaaS and tech — follow me to stay ahead