ON THE RADAR

Ollama v0.40.14: OpenAI-compatible endpoints keep model sampler defaults

The November 3 fixes release stops overriding packaged sampler defaults — clients that omit temperature/top_p now get the model's own values instead of 1.0 — alongside Claude Code limit, web search and Codex routing fixes.

AI News In FocusUpdated Nov 4, 2026 2 min read
Official source published Nov 3, 2026

Ollama released v0.40.14 on November 3, 2026, a fixes-only release with no new models. Read the release notes.

The change API users should check

OpenAI-compatible endpoints no longer override a model's packaged sampler defaults. Previously, clients that did not send temperature or top_p effectively got 1.0; they now receive the model's own packaged values, matching what /api/chat already did. Clients that rely on the old implicit 1.0 behavior should start sending the values explicitly.

Other fixes

  • ollama launch claude applies the model's output token limit for Claude Code, so cloud models no longer hit the 128k default cap.
  • Cloud web search responses on /api/chat decode correctly, including compressed responses.
  • /v1/responses preserves readable input content, including prior assistant refusals, instead of flattening it.
  • The desktop app retries ChatGPT disconnect failures instead of leaving the connection broken; ollama launch codex verifies ChatGPT routing before configuring and reports a clear error when unavailable.
  • Top-k sampling on MLX is faster for large vocabularies.

Recent versions: v0.40.13 and v0.40.12.

Original sourceRead the official source
More AI news Suggest a correction