MiniMax's full model family lands on Amazon Bedrock, including new agent-tuned M2.5

Amazon Web Services has added MiniMax's open-weight model family to Amazon Bedrock, giving AWS customers direct, managed access to MiniMax M2, M2.1, and the newly released M2.5 alongside the rest of Bedrock's model catalog.

What's new

AWS confirmed the addition in a Bedrock machine-learning blog post published July 6, 2026: "The MiniMax family is available on Amazon Bedrock, giving you three open-weight models to match different production workloads." The three models are exposed under the model IDs minimax.minimax-m2, minimax.minimax-m2.1, and minimax.minimax-m2.5.

M2.5 is the headline addition — AWS calls it "the newest model available on Amazon Bedrock" and says it was built "for agent-native execution" through "training that emphasizes tool-calling, multi-step task decomposition, and long-horizon coding tasks." That positions it against other coding- and agent-focused open models rather than as a general chat assistant.

Availability is broad out of the gate: the models are live in 14 AWS regions, including US East (N. Virginia), US West (Oregon), Europe (Frankfurt), and Asia Pacific (Tokyo). Customers can reach the models through either the bedrock-runtime or bedrock-mantle endpoints, and all three support Bedrock's Standard, Priority, and Flex service tiers, letting teams trade latency for cost depending on the workload.

AWS also addressed data handling directly, a recurring question for third-party models hosted on Bedrock: "Your prompts and completions are not used to train any models, and your content is not shared with the model providers." That guarantee mirrors the data-isolation terms AWS already offers for other non-Amazon foundation models on the platform.

Context

MiniMax has been building out an independent presence among open-weight labs, with M2-series releases aimed squarely at coding and agentic use cases rather than trying to match closed frontier chat models on general benchmarks. Landing all three generations on Bedrock in one move — rather than trickling in a single model — suggests AWS is treating MiniMax as an established, ongoing catalog partner rather than a one-off addition.

For AWS, this continues a pattern of diversifying Bedrock's non-Amazon model roster beyond the usual closed-model partners, giving enterprise customers more open-weight options they can run under the same IAM, billing, and compliance controls they already use for Anthropic or Meta models on the platform.

Why it matters

For enterprises already standardized on Bedrock, this removes the operational friction of self-hosting MiniMax's weights just to get agent-tuned coding performance — they can now provision M2.5 the same way they provision any other Bedrock model, with the same regional footprint and data guarantees. It's also a signal of where competitive pressure is coming from: a Chinese open-weight lab's agent-specific model is now distributed through the same managed channel as frontier U.S. labs' models, on equal technical footing inside AWS's own infrastructure.