The Pulse
Ollama Adds Model Thinking Metadata to Its API
Ollama’s v0.34.3-rc1 pre-release adds thinking controls and defaults to the metadata returned by GET /api/show, alongside three other changes.
AI.info Team ·
Three settings now appear in Ollama’s model metadata
Ollama’s v0.34.3-rc1 release candidate adds a way for applications to inspect the thinking controls advertised by an individual model. The release notes state that GET /api/show now advertises each model’s thinking controls and default.
The GitHub release page labels the entry “Pre-release.” It shows the release as v0.34.3, identifies the associated tag as v0.34.3-rc1, and lists the release date as 19 Sep. The page does not specify a year in the displayed date. The release is associated with commit 6383a0f, which GitHub marks as created on GitHub.com and signed with GitHub’s verified signature.
The example in the release notes sends a request to http://localhost:11434/api/show with the model name glm-5.3-flash:cloud:
curl http://localhost:11434/api/show -d '{"model": "glm-5.3-flash:cloud"}'The response shown on the page contains a thinking object. It lists three available values — low, high and max — and identifies max as the default:
{ "thinking": { "values": ["low", "high", "max"], "default": "max" }
}The documented behavior concerns model metadata returned by the endpoint. A client can query the selected model and inspect the controls that model advertises. The example does not establish that every Ollama model supports the same values or uses the same default.
Three other changes in the release candidate
The release notes list support for Nemotron H vision models on Apple Silicon with MLX. The entry does not provide additional details about supported hardware configurations, performance, installation requirements or limitations.
The release also changes the behavior of Ollama’s macOS application. The notes say the app will no longer reopen windows that users have closed when they activate it. The release page does not describe how activation is handled internally or whether the behavior applies to every type of previously closed window.
A third listed change is a fix for model pulls from HuggingFace. The release note does not identify the failure mode, affected repositories, error conditions or the exact correction. Its practical effect therefore cannot be determined from the release entry alone.
What the release page confirms
The page links the full changelog to a comparison between v0.34.2 and v0.34.3-rc0. It lists 19 release assets, although the page content does not provide further details about those assets in the release summary.
The API change described on the page is limited to the thinking metadata returned by /api/show. The release notes do not describe changes to chat or generate responses, specify additional thinking fields, identify other supported model families, or state defaults beyond the example for glm-5.3-flash:cloud. Those details are not verified by this release announcement.
The release page contains repository release notes and no statement attributed to a named person. No attributable quotation is available; the page was checked for a speaker, role and quoted announcement, but it contains only the release description and bullet-point changelog.