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

Amazon Bedrock Reinforcement Fine-Tuning Updates

Amazon Bedrock now supports reinforcement fine-tuning (RFT) for popular open-weight models, including OpenAI GPT-OSS and Qwen models, and introduces OpenAI-compatible fine-tuning APIs. This update simplifies improving model accuracy for developers without requiring extensive machine learning expertise or large datasets.

Reinforcement fine-tuning automates the customization workflow, allowing models to learn from feedback on multiple responses using a small set of prompts. This approach enables the use of smaller, faster, and more cost-effective model variants while maintaining high quality.

Key Features

• Automated end-to-end customization workflow
• Smaller, faster, and more cost-effective model variants
• High-quality model customization

Models Supported

• qwen.qwen3-32b
• openai.gpt-oss-20b

What to do

• Define reward functions using rule-based graders or AI-based judges
• Use AWS Lambda functions for custom grading logic
• Access intermediate model checkpoints for evaluation and debugging
• Use fine-tuned models for on-demand inference through Amazon Bedrock’s APIs

Source: AWS release notes

If you need further guidance on AWS, our experts are available at AWS@westloop.io. You may also reach us by submitting the Contact Us form.