ollama releases

Details

Key Features and Improvements

ollama launch claude: Applied the model's output token limit to Claude Code, resolving the issue where cloud models reached the 128k default limit.
/v1/responses: Improved to maintain readable input content, including the refusal content of previous assistants.
MLX performance: Improved top-k sampling speed when using large vocabularies.

Bug Fixes

/api/chat: Fixed an issue so that cloud web search responses (including compressed responses) are decoded correctly.
OpenAI-compatible endpoints: Fixed an issue where the value was overwritten to 1.0 when the client did not send temperature/top_p; now, the default sampler values from the model package are applied identically to /api/chat.
Ollama app: Improved to allow retries without leaving the connection status as broken when a ChatGPT connection fails.
ollama launch codex: Fixed to validate ChatGPT routing before configuration and report a clear error if routing is unavailable.

Key Features and Improvements

ollama launch claude: Applied the model's output token limit to Claude Code, resolving the issue where cloud models reached the 128k default limit.
/v1/responses: Improved to maintain readable input content, including the refusal content of previous assistants.
MLX performance: Improved top-k sampling speed when using large vocabularies.

Bug Fixes

/api/chat: Fixed an issue so that cloud web search responses (including compressed responses) are decoded correctly.
OpenAI-compatible endpoints: Fixed an issue where the value was overwritten to 1.0 when the client did not send temperature/top_p; now, the default sampler values from the model package are applied identically to /api/chat.
Ollama app: Improved to allow retries without leaving the connection status as broken when a ChatGPT connection fails.
ollama launch codex: Fixed to validate ChatGPT routing before configuration and report a clear error if routing is unavailable.

ollama