← Back to digest
Top Highlight

MiniMax-H3 goes public hours after Qwen3.8-Max

What's this about?

Chinese lab MiniMax made MiniMax-H3 publicly available, with weights on HuggingFace — landing within hours of Alibaba's Qwen3.8-Max announcement. simonw's dry observation captured the pace: "Qwen 3.8 Max and MiniMax-H3 within hours of each other."

What exactly is MiniMax?

MiniMax is one of China's "AI tiger" startups (behind the Hailuo video models and earlier MiniMax-M/H text models), known for aggressive open-weight releases. H3 is its latest flagship language model. The pattern to watch: Chinese labs now routinely open-source models at a cadence and quality level that directly pressures US labs' pricing — last week's 80% GPT-5.6 Luna price cut was widely read as a response to exactly this.

Why it matters

Two frontier-adjacent open releases in one day is the "open model wars" in fast-forward: for builders, the local/self-hosted option keeps getting better on a weekly cycle, and the strategic gap between "open weights" and "API-only" keeps narrowing.

← Google had "ChatGPT" a year before ChatGPTSakana AI opens the Namazu API →