Amazon and Cloudflare Open-Source New AI Decision Models
Amazon and Cloudflare have released open-source AI decision models designed to help autonomous agents make faster, more structured choices instead of relying entirely on large language models to determine their next actions.
Amazon Web Services and Cloudflare have both entered the growing market for AI decision models, releasing new open-source systems designed specifically to help AI agents decide what to do next. AWS introduced Strands Decider 2B, while Cloudflare released Clef and Clef-flash. The developments come as AI companies increasingly focus on building agents that can operate independently and need fast, reliable mechanisms for choosing actions within predefined workflows.
Cloudflare’s Clef models are designed to handle decision-making rather than generate long-form responses. The models analyze an input and return structured probabilities for predefined choices, allowing an AI agent to select an appropriate action without first producing a lengthy block of text. Cloudflare says Clef has a 64,000-token context window and includes a vision encoder, while the smaller Clef-flash model is optimized for extremely low latency. Both models are available as open weights under the Apache 2.0 licence.
Amazon’s Strands Decider 2B takes a similar approach, providing developers with a relatively small model that can make decisions for agentic applications. AWS says the model is designed for fast experimentation and local development, allowing developers to run it without relying entirely on cloud-based inference. The model is part of Amazon’s Strands ecosystem for building AI agents and is being released as an open-source project.
The releases highlight a shift in AI-agent architecture. Traditional large language models are capable of reasoning, generating text and calling tools, but using a large model for every individual decision can increase latency and computing costs. Decision models can instead handle narrower choices, such as selecting which tool an agent should use or determining the next step in a workflow. Cloudflare says Clef can be combined with larger language models so that the decision model handles routing while the larger model performs more complex reasoning or generation.
The competition is developing rapidly, with Cloudflare’s release arriving shortly after other companies began experimenting with similar decision-focused AI systems. Cloudflare says its models are available through Workers AI, while the weights can also be downloaded for self-hosting. The company is positioning the technology as a building block for faster and more reliable AI agents, potentially allowing developers to create systems that make decisions quickly without repeatedly invoking expensive general-purpose models.
