MiniMax has released the MiniMax M3.1 Flash on its API at a pricing rate of $0.15 per one million input tokens. The company also confirmed that its earlier stealth model, Space Bunny, was an early build of this new release. Designed for developers, the model delivers near-frontier agentic scores at a lower cost tier, with model weights scheduled to arrive in late October 2026.
The new multimodal mixture-of-experts model features 428 billion total parameters with about 23 billion active parameters and supports a one-million-token context window. It handles text, image, and video inputs, offering always-on reasoning with five adjustable effort levels. Available now through the MiniMax API, M Plan, and MiniMax Code, the model is priced at $0.15 for input and $0.60 for output per one million tokens, with cache reads costing $0.015 per one million tokens.
This release matters because it brings near-frontier agentic and coding performance to developers at a Flash-tier cost of $0.15 per million input tokens, significantly lowering the expense of running large-scale multimodal workflows.