Chinese AI startup MiniMax has launched its M3 large language model, and the benchmark numbers are turning heads: frontier-level performance at roughly 5 to 10 percent of what leading U.S. models charge per token.
Released Sunday evening, MiniMax-M3 was benchmarked against GPT-5.5 and Gemini 3.1 Pro on tasks spanning coding, reasoning, and multi-step agentic workflows. Results published with the launch show M3 matching or exceeding both models on several core benchmarks while carrying a price of $0.60 per million input tokens and $2.40 per million output tokens, compared to $10 to $15 or more per million tokens for comparable tiers of leading U.S. offerings.
The cost differential is significant enough that enterprise procurement teams will need to take it seriously. For AI-intensive applications such as content pipelines, code generation at scale, or autonomous agents processing high message volumes, a 10x to 20x cost reduction can fundamentally change how organizations budget for AI. MiniMax is also releasing M3 under an open-source license with open weights, meaning enterprises can download, fine-tune, and self-host the model without per-token charges or API dependency.
MiniMax characterized M3 as evidence that next-generation AI performance improvements will come from architectural efficiency rather than simply training larger models on more data. This mirrors a broader industry trend where models like DeepSeek-R2 have demonstrated that raw scale is not the only path to frontier capability.
Why It Matters
MiniMax-M3 intensifies competitive pressure on every commercial AI provider and accelerates the commoditization of frontier AI capability. For organizations evaluating AI infrastructure strategy, M3's combination of open weights, competitive benchmarks, and dramatically lower operating costs makes it a serious candidate for deployment in latency-tolerant, high-volume workloads. The open-weights license also enables private cloud or air-gapped deployments where API-based models are not viable. Expect major U.S. AI providers to respond with pricing adjustments or new efficiency tiers in the coming weeks.
MiniMax has positioned M3 as a foundation model for autonomous agent development, with tooling and API compatibility designed to ease migration from GPT and Gemini-based stacks.