MiniMax M2.5

MiniMax M2.5, released in February 2026, is an open model from Chinese AI startup MiniMax that delivers frontier-level performance at dramatically lower costs. It scores 80.2% on SWE-Bench Verified for software engineering, outperforming Claude Opus 4.6, GPT-5.2, and Gemini 3 Pro on programming tasks. On web search (BrowseComp: 76.3%) and tool use (BFCL: 76.8%), it also leads the field.

The lightweight M2.5-Lightning variant generates 100 tokens per second, with MiniMax claiming one hour of continuous operation costs just one dollar. This positions M2.5 at roughly 1/20th the cost of Claude Opus 4.6 while matching or exceeding its performance. The predecessor MiniMax M2, released in October 2025, had already placed among the top five models globally on the Artificial Analysis intelligence index.

MiniMax completed its Hong Kong IPO in January 2026, raising $619 million with shares surging 109% on debut (briefly valued at ~$13 billion). The company's promise of "intelligence too cheap to meter" reflects the broader trend of Chinese AI labs aggressively undercutting Western pricing while maintaining competitive performance, creating significant pressure on the business models of established AI providers.