Ollama Adds Speech and Decision Models in v0.40.11

What happened

Ollama has added speech generation capabilities on macOS, allowing users to hear text or chat model responses as they are generated. The feature is accessible via the 'ollama run' command, which plays audio in real time.

The platform also integrates d3 decision models, which use vLLM Semantic Router's decision models for classification, yes/no answers, and probability scoring. These models are available in sizes of 2B, 4B, 9B, and 27B parameters.

Why it matters

The speech generation feature enhances user interaction by providing immediate auditory feedback, while the d3 models offer structured decision-making capabilities for tasks like customer issue classification. These updates expand Ollama's utility in enterprise and developer workflows.

Key details

The d3 models can be queried with text, structured state, or images to receive classifications or responses. Applications can also stream audio via the OpenAI-compatible '/v1/audio/speech' endpoint. A new contributor, SynapShift, made their first contribution through pull request #19173.

The release includes a full changelog comparing v0.40.10 and v0.40.11-rc0. The speech generation feature includes an example command that plays the text 'Hello from Ollama.' aloud.