Amazon Bedrock reinforcement fine-tuning adds support for open-weight models with OpenAI-compatible APIs

This article announces Amazon Bedrock's expansion of reinforcement fine-tuning (RFT) support to open-weight models with OpenAI-compatible APIs.

• RFT now supports GPT-OSS-20B and Qwen3 32B open-weight models
• Enables model improvement using feedback on multiple responses, not large labeled datasets
• Customers define reward functions via rule-based graders or AI-based judges
• AWS Lambda functions support custom grading logic during training
• Access intermediate checkpoints for model evaluation and selection
• Fine-tuned models deploy immediately via OpenAI-compatible APIs
• All proprietary data remains secure within AWS environment

Amazon Bedrock simplifies foundation model customization for business needs without requiring ML expertise or complex infrastructure.