minimax-m3

MiniMax M3 is MiniMax's flagship model, succeeding the M2.7 series with substantially improved reasoning capability, vision support, and a 1M-token context window. M3 introduces chain-of-thought reasoning to the MiniMax lineup for the first time, achieving an intelligence score of approximately 84/100 — placing it in the upper frontier tier. M3 represents MiniMax's most capable model to date.

MiniMax M3 is priced at $0.80 per million input tokens and $2.40 per million output tokens, with prompt caching at $0.40/1M — making it one of the most cost-effective frontier-tier reasoning models available. At 1,000 input + 500 output tokens per call, it costs approximately $0.0020 per 1,000 API calls — significantly cheaper than comparable frontier-tier reasoning models. Throughput is approximately 130 tokens per second with a median TTFT of 340ms — fast for a reasoning model. The 1M-token context window is among the largest available at this price point.

MiniMax M3 supports vision inputs, function calling, streaming, JSON schema output, and audio input. Its benchmark scores reflect strong frontier-tier performance across reasoning, coding, and multimodal tasks. The chain-of-thought reasoning approach improves performance on multi-step math, complex coding, and agentic tasks compared to M2.7. Prompt caching and audio input support make it suitable for voice-enabled and multi-turn applications.

MiniMax M3 is best suited to cost-sensitive teams requiring frontier-tier reasoning capability, long-context processing, or multilingual support. Its $0.80/1M input price makes it one of the most affordable frontier reasoning models available — significantly cheaper than comparable reasoning models from other providers. For teams building agentic workflows, RAG pipelines, or long-document analysis tools where cost scales with usage, M3's combination of frontier-tier intelligence and budget-friendly pricing is compelling. Audio input support enables voice-to-text-to-reasoning pipelines without additional transcription costs.

MiniMax M3 sits above MiniMax M2.7 ($0.30/1M input, intelligenceScore 75, no reasoning) and MiniMax M1 ($0.30/1M input, intelligenceScore 74, reasoning) in the MiniMax lineup. Its main competitors are frontier-tier reasoning models from OpenAI and efficient-tier models from Google, all priced differently. M3's combination of frontier-tier intelligence score (84) and sub-$1/1M input pricing is unusual in the market. Its intelligence score of approximately 84/100 reflects genuine frontier-tier performance validated across multiple benchmark categories.