OllamaFrom the builder
Ollama's highest performance on Apple Silicon yet with MLX
Ollama updated its MLX engine for Apple Silicon, adding support for NVIDIA's NVFP4 format and a new snapshot system that cuts repeated work in agent sessions. On Gemma 4 12B, NVFP4 roughly halves the quality loss of the common q4_K_M quantization while keeping performance. Fusing several operations into single Metal kernels and reworking GPU sampling makes the engine up to 20% faster.
Agent workloads are dominated by prompt processing. Every tool call is a new request, and every request resends the whole transcript: system prompt, tool definitions, and every file read so far.
Saved . A shortened excerpt from the official blog; the original publisher retains authorship.