MiniMax Launches M3 Model With 1M Context and Native Multimodal Capabilities

MiniMax (Shanghai Hixi Technology) launched its M3 model on June 1, 2026, positioning it as the first domestic AI model to integrate frontier coding, agentic capabilities, million-token context windows, and native multimodal processing — all within a single architecture. Built on MiniMax's proprietary Sparse Attention (MSA) architecture, the M3 API supports up to 1 million tokens of context, with a guaranteed minimum of 512K tokens available. This makes it suitable for long-range agent tasks, extended coding sessions, and long-form video understanding. The model achieves "industry-leading" performance in coding and agent benchmarks, with autonomous task decomposition, tool invocation, and multi-step reasoning capabilities. The M3 is a natively multimodal model, having been trained on multimodal data from the ground up. MiniMax restructured its entire data pipeline, scaling pre-training data to hundreds of terabytes and achieving tight alignment between text and visual semantic spaces. In the BrowseComp agent benchmark, M3 scored 83.5, surpassing OpenAI's Opus 4.7 (79.3). In a demonstration of autonomous capabilities, MiniMax tasked the M3 with reproducing an ICLR 2025 outstanding paper on the learning dynamics of LLM fine-tuning. The model ran for nearly 12 hours independently, producing 18 commits and 23 experimental charts, successfully executing the core experiments. MiniMax also tested M3's ability to function as an AI research assistant, giving it four pre-trained base models and instructing it to complete data synthesis, training, evaluation, and iteration within 12 hours — all without human intervention. The M3 scored 37.1, ranking third behind Opus 4.7 (42.4) and GPT-5.5 (39.3). The model is available in two API versions — M3 and M3-highspeed — with identical results but faster inference on the latter. Automatic caching is supported and enabled by default. MiniMax plans to open-source M3 on HuggingFace and GitHub, supporting private cluster deployment and fine-tuning. Pricing for the M3 API (context ≤512K) is offered at a 50% discount for the first seven days: input at 2.1 yuan per million tokens (standard) or 3.15 yuan (priority), and output at 8.4 yuan (standard) or 12.6 yuan (priority). Cache reads cost 0.42 yuan (standard) or 0.63 yuan (priority) per million tokens.