Ollama 0.40.9 landed this week, and it follows 0.40.8’s Flint release.
Highlights
- Speech: Ollama can now generate speech with Phonon on Apple Silicon.
ollama runplays text or a chat model’s reply aloud as the audio is generated, and apps can stream audio through the OpenAI-compatible/v1/audio/speechendpoint. - d3 decision models: these bring vLLM Semantic Router’s decision models to Ollama. You give a model text, structured state or an image and ask it to classify, answer yes or no, or score options with probabilities. The family comes in 2B, 4B, 9B and 27B sizes. The release notes’ example asks whether “A customer says their card was charged twice” is a billing issue.
- Flint on Linux and Windows: the coding model from 0.40.8 now runs on Linux and Windows as well as macOS.
Other improvements
- Streaming OpenAI Responses and Anthropic Messages requests now report generation failures instead of ending silently after partial output.
ollama launchnow passes cloud models’ context lengths and capabilities to coding tools, fixing incorrect context and compaction settings.- Anthropic-compatible web search now uses the model’s supported thinking setting in follow-up requests, preventing avoidable failures.
My take
The d3 models are the sleeper here. A small local model that answers “is this X?” with probabilities is exactly what you want sitting in front of an agent as a router or a guardrail. And local text-to-speech through an OpenAI-style endpoint means existing apps can talk with almost no changes, as long as you’re on Apple Silicon for now. I put together a short how-to for trying both locally.
Have a blessed day! And Happy Hacking!
