Run MiniMax models on Amazon Bedrock

This article announces the availability of MiniMax M2 family models on Amazon Bedrock, a fully managed service for accessing frontier foundation models with data protection and compliance guarantees.

• Three MiniMax models available: M2 (1M token context), M2.1 (improved reasoning), M2.5 (agent-native, RL-trained)
• Access via bedrock-mantle endpoint (Chat Completions API, OpenAI SDK compatible) or bedrock-runtime (AWS SDK)
• Tool-calling support for agentic workflows with client-side function invocation
• Service tiers: Standard (on-demand), Priority (25% better latency), Flex (cost-effective, higher latency)
• Implicit prompt caching reduces latency for repeated prefixes across requests
• Scaling best practices: exponential backoff for 503 errors, gradual traffic ramps, Priority tier for latency-sensitive workloads
• Available in 14 AWS Regions with per-token pricing

MiniMax models enable production AI workloads including agentic applications, long-context analysis, and software engineering tasks on AWS infrastructure with full data privacy.