MiniMax M3

MiniMax M3 is an open-weights multimodal model from Chinese AI lab MiniMax, released on June 1, 2026. It competes directly with frontier models like GPT-5.5 and Gemini 3.1 Pro on coding and agentic benchmarks — the key difference being that it does so at roughly 5-10% of the cost. The model supports a 1M token context window natively, handles text, image, and code inputs, and is designed from the ground up for agentic workflows where the model needs to plan, use tools, and execute multi-step tasks.

What makes M3 genuinely interesting for developers is the combination of price and capability. API pricing sits at $0.30 per million input tokens and $1.20 per million output tokens during the current promotional period — dramatically cheaper than comparable closed models. The 1M context window is large enough to ingest entire codebases, long document chains, or extended conversation histories without chunking. Open weights mean you can self-host, fine-tune, and inspect the model, which matters for teams with data sovereignty requirements or specialized use cases.

The main caveats are ecosystem maturity and geographic origin. MiniMax's developer platform and tooling are less polished than OpenAI's or Anthropic's, and third-party integrations (IDE plugins, agent frameworks) are still catching up. As with DeepSeek, the Chinese company origin may raise compliance questions for some enterprise buyers. But on raw price-performance for coding and agentic tasks, M3 is hard to ignore — it's already available on OpenRouter and other inference providers, which lowers the adoption barrier considerably.