Ollama 0.40.14 Enhances Cloud Model Limits and Endpoint…

Ollama 0.40.14 Enhances Cloud Model Limits and Endpoint Behavior

Ollama 0.40.14 improves cloud model token limits, web search decoding, and OpenAI-compatible endpoint defaults.

By Precis Daily NewsroomNovember 5, 20261 min read174 words

AI-summarized from the linked source and published automatically — not human-edited.How this works →

What happened

Ollama's v0.40.14 update ensures the 'claude' command respects the model's output token limit, preventing cloud models from exceeding the 128k default cap. This change avoids potential issues with token overflow in large responses.

The release also fixes decoding issues for compressed web search responses on the /api/chat endpoint, ensuring accurate retrieval of search results. Additionally, OpenAI-compatible endpoints now use model-specific sampler settings instead of defaulting to 1.0 for parameters like temperature or top_p.

Why it matters

These updates improve the reliability and accuracy of cloud-based model interactions, especially for applications relying on web searches or OpenAI-compatible APIs. By aligning endpoint behavior with model defaults, developers can avoid unexpected output configurations.

Key details

The update includes optimizations for MLX-based top-k sampling, enhancing performance with large vocabularies. It also introduces a full changelog detailing changes from v0.40.13 to v0.40.14, providing transparency for users.

Ollama's application now retries ChatGPT disconnect failures, improving connection stability. The 'codex' command now checks ChatGPT routing availability before configuration, offering clearer error messages if routing is unavailable.