Ollama: MLX Performance Push and Compatibility Fixes

MLX sampling gets faster and safer while new JEV and MiMo models land, and the server tightens OpenAI compatibility, media validation, and memory estimates.

Duration: PT2M23S

Episode overview

This episode is a short developer briefing from Ollama.

It explains recent repository work in plain language.

  • Show: Ollama
  • Published: 2026-10-30T13:01:22Z
  • Audio duration: PT2M23S

Transcript excerpt

This excerpt keeps the crawler page concise. Listen to the episode or use the RSS feed for the full update.

Hello, it's October 30th, 2026, and here's what's moving in Ollama.

The lead pattern is maturity: the MLX stack is getting faster and better guarded, while core server logic is closing small correctness gaps that affect reliability and trust.

First, MLX performance and control. A merged change in PR 19392 reworks how large top K selections are sorted, processing long vocabulary rows in chunks instead of a full multi-pass sort. That cuts a major cost for very large vocabularies and applies to sampling and candidate selection for speculative decoding.…

Second, correctness in the server and compatibility layer. PR 19401 fixes OpenAI-compatible response IDs that were drawn from a very small random range, sharply raising collision risk for chat, completions, and responses endpoints. PR 19397 adds a media capability check, so for example an audio clip sent as an image…

Finally, coverage. PR 19399 adds JEV text and vision support, and PR 19400 adds MiMo version 2.6, both with candidate models available for testing. Three community integrations were also listed: Lirnyk, Clairvoyance, and RedAmon.

What's next: if you run MLX or large vocab models, expect lower sampling overhead, and test the…

Nearby episodes from Ollama

  1. Inference Correctness and Installer Reliability
  2. Weekly Recap - Sampler Fidelity, Faster Drafting, and Launcher Fixes
  3. Sampler Defaults Fix and Memory Work
  4. MLX Memory Safety and Silent Correctness Fixes
  5. OpenAI Fidelity and Runner Stability
  6. LFM3 Speech, Video and Reliability Fixes
  7. Scheduler Memory Fixes and Faster Drafting
  8. Mac Prefill Gains and Scheduler Fixes