crewAI 1.15.28 Adds 'eval' Command and Context Window Enhancements

What changed?

crewAI version 1.15.28 introduces the 'crewai eval' command, which reports where a model failed after being stopped. The update switches context window handling to use the LiteLLM catalog, which may expand model compatibility and accuracy. Other fixes include improved handling of human feedback, better error messaging in deployments, image input support for certain OpenAI GPT-6 models, and event emission for streamed Anthropic tool calls.

Why does it matter to an everyday developer?

The 'crewai eval' command makes it easier to diagnose and fix workflow failures in model deployments, reducing debugging time. Integration with the LiteLLM catalog standardizes and potentially improves model context window configurations, which affects prompt size limits and reduces unexpected errors. Improved handling of human feedback, clearer error reporting, and expanded support for image inputs in newer OpenAI models directly benefit developers building or maintaining agent workflows using crewAI.

What can the developer do now?

Developers can upgrade to crewAI 1.15.28 to use the new 'crewai eval' diagnostic tool to quickly identify model failures. They should review the updated context window handling to ensure prompt sizing behaves as expected across models. Developers using OpenAI GPT-6 Sol, Luna, or Astra can now pass image files as inputs, opening up new multimodal agent capabilities.