Ollama Releases Pre-Release Candidate v0.32.6-rc0 on Aug 4, 2026 | ai-inference
Ollama released pre release version v0.32.6 rc0 on Aug 4, 2026 as a material update to its AI inference runtime and model server, featuring MLX and llama.cpp e…
Published on MyPrivateClaw
Aug 4, 2026, 11:12 PM UTC
Coverage date
Aug 4, 2026
Last updated
Aug 4, 2026, 11:12 PM UTC
News summary
The MLX engine in v0.32.6 rc0 now uses the model's MTP head for speculative decoding automatically, improving Qwen3.5 speed on Apple GPUs. /v1/chat/completions streaming in v0.32.6 rc0 now matches OpenAI's wire format: role only on the first chunk, finish reason on its own chunk, and usage in a separate chunk with stream options.include usage. Truncated OpenAI responses in v0.32.6 rc0 now report finish reason "length" instead of "tool calls". ollama run kimi k3 in v0.32.6 rc0 now offers kimi k3:cloud for cloud only models that publish no default tag, instead of failing. Experimental image generation has been temporarily removed in v0.32.6 rc0; users should continue using 0.32.5 for image generation support. TUI fixes in v0.32.6 rc0 include: pipe delimited prose no longer renders as a table, Enter accepts the highlighted @ file completion, and /prompt scrolling is no longer laggy.