v0.21.3

Ollama has shipped v0.21.3, a pre-release version that introduces two key API improvements. The first, contributed by ParthSareen, allows the think parameter to accept a 'max' value, enabling models to engage in deeper reasoning before generating responses. This is particularly useful for complex tasks requiring more cognitive effort, though it may increase latency. The second change maps OpenAI's 'reasoning_effort' parameter to Ollama's think system, making it easier for developers to switch between platforms without rewriting code.

These updates are part of Ollama's ongoing effort to bridge local and cloud-based AI workflows. By aligning with OpenAI's parameter naming, Ollama reduces friction for developers experimenting with local models. The 'max' think value also hints at future optimizations for reasoning-heavy applications like code generation or multi-step analysis. While still in pre-release (v0.21.3-rc0), the changes signal Ollama's commitment to flexibility and interoperability in the open-source AI ecosystem.