BriefPulse Practical AI · Working notes on AI you can actually use. RSS · BriefPulse network
BriefPulse Practical AI

What changed in AI, what it is useful for, and what you can do with it.

17 September 2026

Brief

Ollama 0.31 speeds up Gemma 4 on Apple Silicon with MLX multi-token prediction

Ollama says Gemma 4 runs up to 90% faster on Apple Silicon in version 0.31 using MLX-backed multi-token prediction, with the figure scoped to coding agents. The post names the Aider polyglot benchmark but does not set out its method or raw results, so treat the number as vendor-reported.

Photograph
Photograph: Dbirdz · CC BY-SA 4.0 source

Ollama 0.31 ships Gemma 4 with multi-token prediction on Apple Silicon, a speed-up the company attributes to MLX. Multi-token prediction has a model propose more than one token at a time, which cuts the number of sequential decode steps. On a Mac that is usually the step that limits throughput, because memory bandwidth rather than raw compute tends to be the binding constraint.

The headline figure is scoped to a single workload: up to 90% faster when Gemma 4 is used with coding agents, measured with the Aider polyglot benchmark. Nothing else about the model's access, interface or cost changes for readers here; the difference is how quickly it decodes on Apple hardware.

Our reading

Local decode speed is the main reason to run a coding agent on a laptop instead of calling an API, so a large speed-up on Apple Silicon is a practical change for anyone working offline, on metered connections, or trying to keep code off third-party servers. The 90% figure is vendor-reported and tied to one benchmark and one workload, which is a reason to reproduce it rather than assume it transfe…

What to do or watch

Run a representative coding task, or the Aider polyglot benchmark itself, on Ollama 0.31 and on your previous version on the same Mac, and record wall-clock time per task before relying on the 90% number.

Source details and supporting facts

Each line is stated by the page named above it.

Stated by ollama.com

  • Gemma 4 is now significantly faster in Ollama 0.31 on Apple Silicon via multi-token prediction (MTP), powered by MLX.
  • Performance is now up to 90% faster when used with coding agents, as measured using the Aider polyglot benchmark.

Sources

  1. Ollama blogText stored 17 September 2026

How this story was checked. Written from the 1 page listed above, stored 17 September 2026; claims checked against that stored text on 17 September 2026.

What that means
  • 2 of 2 reported statements were confirmed against the page that carries them; the rest were removed rather than published.
  • Figures in the text were required to appear in the stored source text: yes. Identifiers: yes.
  • The check reads stored text only: no claim rests on a fresh look that did not happen.
  • Where the reporting was silent, the text says so instead of filling the gap.

More from Practical AI