Amazon Bedrock hits nearly 100 models with 18 new open-weight additions

Amazon Bedrock just crossed a threshold: nearly 100 serverless models, all fully managed, all accessible through a single API.

Today's drop includes 18 new open-weight models. We're talking Mistral Large 3 (multimodal, long-context beast), the Ministral 3 family (3B, 8B, 14B for edge to enterprise), Google Gemma 3 (4B to 27B), NVIDIA Nemotron, Qwen3, Moonshot AI's Kimi K2 Thinking, MiniMax M2 for coding agents, and OpenAI's safety models.

Here's what matters: you don't have to pick once and live with it forever. Test, evaluate, switch, deploy—all without rewriting your application or changing infrastructure. One unified API across everything.

For DevOps and platform teams, this is massive. Edge deployment? Ministral 3 3B. Complex reasoning? Mistral Large 3 or Kimi K2. Coding automation? MiniMax M2. Privacy-critical? Deploy locally with Gemma. RAG pipelines? Qwen3 handles ultra-long documents.

The real story isn't the number of models. It's the flexibility. You're not locked in. You can test in the Bedrock console, use any AWS SDK (InvokeModel, Converse APIs), integrate with agentic frameworks, or deploy with Bedrock Agent Core. Guardrails and model evaluation tools are built in.

Check region availability, start testing. The barrier to trying new models just got a lot lower.