INFORUNTIME UPDATE·2026-10-07
Ollama 0.34 and 0.35: ChatGPT Desktop, stable MLX create, decision models
▼ WHAT HAPPENED
Ollama shipped a run of releases in September 2026. 0.34.0 (Sep 5) lets Ollama models be used directly in ChatGPT Desktop, with setup from the Ollama app on macOS. 0.34.1 (Sep 14) made MLX safetensors `ollama create` no longer experimental, and GGUF model creation now requires llama.cpp tooling for safetensors conversion and quantization. 0.34.3 (Sep 19) has `GET /api/show` advertise each model's thinking controls, and 0.34.4 (Sep 23) applies structured outputs on thinking models in a single pass. 0.35.0 (Sep 28) adds decision models through `/v1/systemone`, which return choices, probabilities and scores instead of text. A pre-release, 0.40.0-rc0 (Sep 25), runs MLX-supported architectures on MLX by default on Apple Silicon.
▼ OPERATOR ANGLE
If you build GGUF models with `ollama create` from safetensors, move conversion and quantization to llama.cpp tooling before upgrading past 0.34.1. On Apple Silicon, try the 0.40 release candidate on a non-critical machine first: MLX by default changes memory behaviour and throughput for supported architectures, so re-measure before rolling it out.
▼ ENTITIES REFERENCED
[pulse item] · runlocalai.co/pulse/ollama-0-34-0-35-chatgpt-desktop-mlx-create-decision-models