MiniMax introduces Music 3.0, a next-generation music generation model that composes, arranges, performs, and produces a complete song from a creative concept and optional lyrics in a single…
MiniMax launches H3, a multimodal generation model that understands and generates text, images, video, and audio with native stereo sound. It achieves up to 15 seconds of 2K video.
MiniMax's M3 model, using the MaxProof framework, exceeded the human gold-medal threshold on IMO 2025 and USAMO 2026 benchmarks.
MiniMax announces M3, a new model that reaches frontier capability on coding and agentic tasks, introduces MiniMax Sparse Attention (MSA) supporting up to 1M context, and is natively multimodal.
MiniMax announces an overall upgrade of its Agent, now named Mavis (MiniMax as a Jarvis), an AI butler designed for long-running tasks and continuous evolution.
MiniMax published an internal investigation into the sparse token forgetting phenomenon affecting the MiniMax M2 series of LLMs.
MiniMax announces M2.7, its first model that participates in its own evolution, with native Agent Teams and self-evolving capabilities, excelling in software engineering, professional work, and…
The blog post introduces Forge, a scalable reinforcement learning framework and algorithm designed to address the trade-off between system throughput, training stability, and agent flexibility for…
MiniMax M2.5 is built for real-world productivity.
MiniMax describes how they built a role-play agent for production usage, optimized over three years in Talkie/Xingye.
MiniMax M2.1 is an enhanced model with industry-leading multi-language programming capabilities across Rust, Java, Go, C++, TypeScript and more, featuring enhanced mobile development, aesthetic web…
MiniMax launched and open-sourced the M2 model, an agent-first model that delivers top-tier coding, tool use, and deep search at 8% of Claude Sonnet's price with 2x speed.