MiniMax-M3 tops Next.js agent evaluations

The rise of MiniMax-M3 demonstrates that the gap between open-weight and proprietary frontier models is rapidly shrinking in specialized domains like agentic coding. By optimizing for sparse attention and long-context reasoning, MiniMax has delivered proprietary-grade software engineering capabilities at a pricing tier that makes production-scale AI agents economically viable.

• –High-Efficiency Architecture: The use of MiniMax Sparse Attention (MSA) enables a massive 1-million-token context window while dramatically slashing compute and inference costs.
• –Economically Disruptive Pricing: At 10x cheaper standard (and 20x cheaper via AI Gateway), MiniMax-M3 challenges the dominance of expensive APIs like Claude 3 Opus and GPT-5 for developer tooling.
• –Benchmark Leadership: Leading the Next.js agent evaluations positions MiniMax-M3 as a go-to backend for developers building next-generation web dev agents and Vercel AI SDK applications.
• –Open-Weight Competitiveness: Offering open weights for a highly capable, natively multimodal agentic model will accelerate community integrations and custom finetuning.