Ollama 0.33 makes a cancelled prefill resumable, and adds a Claude Desktop path

The 21 August 2026 release fixes a class of cache bugs that made local serving quietly expensive: a cancelled prefill now keeps every restore point it crossed, so a retry resumes where it stopped, and resumed prefills no longer record restore points that fail to cover what they claim — a case that on recurrent-layer models forced a request matching 46,000 of 47,000 tokens to reprocess from zero. It also lets Claude Desktop be pointed at Ollama as a third-party gateway provider.

What it means If you self-host, a broken prefill restore is invisible until you look at latency; this is the release that stops a near-complete cache hit from re-running the whole prompt.