LoginSign Up
← Back to all sparks
O

Ollama

AI-ASSISTANTS
Velocity6.3

Run and manage large language models locally on your own machine.

ollama.com ↗

Ollama's v0.40.6 candidates fix proxied registry downloads after Nemotron landed on MLX.

mlxlocal-modelsmodel-supportquantizationusage-credits
◆Current state
Ollama has moved to v0.40.6 release candidates. rc1 makes model pulls work behind proxies when the registry redirects to a CDN, deferring redirect DNS to the proxy, and fixes a crash in blob downloads; rc0 switches the CLI's --ask mode to stable question identifiers. The v0.40.5 candidates before them added Nemotron 3.5 Super on MLX and broader NVFP4 imports, and v0.40.4 added Kev decision scoring and usage-credit spending from launch.
◆Where it's heading
MLX remains where new model families land first, with each point release pairing one model-support addition with maintenance. Download and CLI plumbing is getting attention for corporate networks and scripted use, while the launch command is gradually tied to paid usage credits.
◆Prediction
Expect stables for v0.40.4 through v0.40.6 to consolidate the Nemotron, Kev and proxy fixes. The entries don't show what the next model addition will be.

◆Recent moves

  1. 22h ago

    v0.40.6-rc1 fixes model pulls behind proxies

    Model downloads behind a proxy now defer CDN redirect DNS to the proxy, and a nil-cancel crash in blob downloads is fixed. It is network plumbing between the model-support releases.

    View source ↗
  2. 1d ago

    v0.40.6-rc0 uses stable ids for --ask mode questions

    The CLI's --ask mode switches to stable identifiers for question keys. It is an internal change that opens the v0.40.6 cycle.

    View source ↗
  3. 2d ago

    v0.40.5-rc2 bumps llama.cpp

    A llama.cpp version bump with no user-facing notes, cut the same day as rc1.

    View source ↗
  4. 2d ago

    v0.40.5-rc1: Nemotron 3.5 Super on MLX, NVFP4 import fixes

    rc1 adds Nemotron 3.5 Super to the MLX runner, fixes Nemotron loading and quantization bugs, and accepts per-expert scales in NVFP4 imports. Another release where MLX gains a model family first.

    View source ↗
  5. 3d ago

    Ollama v0.40.4-rc2 lets launch use usage credits without upgrading

    A single change lets the launch command draw on usage credits without requiring a plan upgrade. It is the first entry in this window to touch Ollama's paid usage rather than the local runtime.

    View source ↗
  6. 4d ago

    v0.40.4-rc1: Add Kev decision scoring to MLX (#18976)

    The MLX runner can now load Kev models with their FP32 decision head and run System One scoring, with shared-state caching and Kev-specific limits. After weeks of migration trims and dependency bumps, it is a new model type landing on MLX first, as the multimodal embeddings did in v0.40.0.

    View source ↗