← All tools

Local / Open Source

Ollama

Updates

  1. Feature12d ago· GitHub Releases

    v0.40.17

    Ollama 0.40.17 adds concurrent generation for MLX and mmap support for Vulkan integrated GPUs. It also improves ChatGPT CLI restoration, graceful restarts, and OpenCode thinking settings.

  2. Feature13d ago· GitHub Releases

    v0.40.16

    Ollama v0.40.16 adds OpenAI truncation forwarding, CUDA 13.5 and SM 103 support for Windows ARM64, and Ripple preview support in the MLX runner. It also improves Qwen3.8 Flash Next memory efficiency and fixes Claude Code routing and retired-model errors.

  3. Feature19d ago· GitHub Releases

    v0.40.15

    Ollama v0.40.15 adds support for MiMo V2.6 models. It also changes Linux MLX handling to load bundled CUDA libraries in-process.

  4. Feature20d ago· GitHub Releases

    v0.40.14

    Ollama v0.40.14 fixes OpenAI Responses input handling, retry behavior for ChatGPT disconnects, sampler defaults, and cloud web search decoding. It also improves MLX top-k sorting and applies model output limits when launching Claude Code.

  5. Feature25d ago· GitHub Releases

    v0.40.13

    Ollama v0.40.13 fixes LFM3 GGUF support and llama-server speech functionality. It also updates the bundled llama.cpp version to b11963.

  6. Feature27d ago· GitHub Releases

    v0.40.12

    Ollama 0.40.12 adds Liquid AI’s LFM3 text, vision, audio, and multimodal models, plus 2B d1 decision models for local use. Initial LFM3 and d1 2B support is limited to Apple hardware, alongside faster Apple Silicon prompt processing for several models.

  7. FeatureOct 24· GitHub Releases

    v0.40.11

    Ollama 0.40.11 improves Qwen3-Coder function parameters, OpenCode context allocation, and vLLM-D3 empty-state handling. It also updates llama.cpp and adds Phonon voice catalog support and text normalization.

  8. FeatureOct 23· GitHub Releases

    v0.40.10

    Ollama 0.40.10 adds NVIDIA’s Nemotron 3.5 Bolt E4B, a Gemma 4-based vision-language model for local chat, image understanding, and tool use. It supports a 128K context window and includes a DFlash draft for speculative decoding.

  9. NewsOct 22· GitHub Releases

    v0.40.9

    Ollama 0.40.9 updates its llama.cpp dependency. The excerpt references a comparison against the 0.40.9 release candidate.

  10. FeatureOct 21· GitHub Releases

    v0.40.8

    Ollama 0.40.8 adds Flint, an Apache 2.0-licensed 40B mixture-of-experts coding model with 262K-token context and about 2.5B active parameters per token. Initial release-candidate support is limited to macOS, with other GPUs planned later.

  11. FeatureOct 18· GitHub Releases

    v0.40.7

    Ollama v0.40.7 allows OpenAI Responses effort settings on non-thinking models, adds automatic draft speculative type inference in llama-server, and supports PPLX Decider v1.2 on MLX.

  12. FeatureOct 16· GitHub Releases

    v0.40.6

    Ollama v0.40.6 prevents orphaned runners on Windows, adds GGUF q2_0 tensor and file types, launches Clef in Ollama Cloud, and changes question keys in --ask mode to identifiers.

  13. FeatureOct 14· GitHub Releases

    v0.40.5

    Ollama v0.40.5 ports Metal custom kernels to CUDA and adds support for custom tool calls in OpenAI Responses history. It also updates MLX, documentation, and cloud payment error messages.

  14. FeatureOct 12· GitHub Releases

    v0.40.4

    Ollama 0.40.4 updates llama.cpp for decision models and EmbeddingGemma 2, reuses verified manifests and loaded models, and fixes MLX recurrent state handling. Installed models now appear in the Codex model picker.

  15. FeatureOct 11· GitHub Releases

    v0.40.3

    Ollama v0.40.3 removes warnings when launching Claude and Codex. It also pauses background model upgrades that added per-request overhead in high-volume workloads such as embeddings.

  16. FeatureOct 8· GitHub Releases

    v0.40.2

    Ollama 0.40.2 adds Oxi to its documented community integrations. It also hides duplicate and downgrade guards from server listings.

  17. FeatureOct 7· GitHub Releases

    v0.40.1

    Ollama 0.40.1 adds proxying for cloud usage and balance APIs and removes the account step from CLI onboarding. It also fixes Windows issues involving large model files and symlinks.

  18. FeatureSep 29· GitHub Releases

    v0.35.1

    Ollama v0.35.1 allows up to 10 web searches per response and adds explicit model capability declarations during model creation. It also updates its MLX and llama.cpp dependencies.

  19. FeatureSep 28· GitHub Releases

    v0.35.0

    Ollama 0.35.0 adds decision models through the /v1/systemone endpoint, returning choices, probabilities, and scores instead of generated text. Initial supported models are Bespoke Labs’ Nimble and Together AI’s Tev1.

  20. FeatureSep 25· GitHub Releases

    v0.40.0

    Ollama v0.40.0 enables its MLX runner by default for supported model architectures on Apple Silicon. Additional models will be tested and enabled during the release candidate period.

  21. FeatureSep 23· GitHub Releases

    v0.34.4

    Ollama v0.34.4 fixes intermittent model-not-found errors and improves structured outputs for reasoning models. It also updates llama.cpp and MLX, adds dynamic Gemma 4 image resolution, and accelerates Qwen 3.8 prompt processing on MLX.

  22. FeatureSep 19· GitHub Releases

    v0.34.3

    Ollama v0.34.3 adds model thinking controls and defaults to the GET /api/show response. It also supports Nemotron H vision models on Apple Silicon through MLX and fixes unwanted window reopening in the macOS app.

  23. FeatureSep 15· GitHub Releases

    v0.34.2

    Ollama 0.34.2 updates its llama.cpp dependency. The release notes provide no additional details.

  24. FeatureSep 14· GitHub Releases

    v0.34.1

    Ollama v0.34.1 improves MLX model memory management, cache eviction, and model-loading behavior. It also updates llama.cpp and MLX, raises the token repeat limit, fixes incomplete-result handling, and refreshes parts of the app interface.

  25. IntegrationSep 5· GitHub Releases

    v0.34.0

    Ollama v0.34.0 lets macOS users run Ollama-hosted open models directly in ChatGPT Desktop. It also improves structured outputs on Apple Silicon and adds OpenAI-compatible tool search, response compaction, and image handling in compacted responses.

  26. FeatureSep 2· GitHub Releases

    v0.33.3

    Ollama v0.33.3 adds reporting for cached prompt tokens and honors default parameters defined in GGUF models. It also updates MLX, MLX-C, and llama.cpp dependencies.

  27. FeatureAug 27· GitHub Releases

    v0.33.2

    Ollama 0.33.2 restores system dark mode, keeps proxy requests running when the model catalog changes, and synchronizes macOS app handoff.

  28. FeatureAug 26· GitHub Releases

    v0.33.1

    Ollama 0.33.1 adds Qwen3.8 Flash Next and structured-output support to its MLX runner. It also updates MLX and llama.cpp and mitigates Metal GPU timeouts when loading models from slow storage.

  29. IntegrationAug 21· GitHub Releases

    v0.33.0

    Ollama 0.33.0 adds Claude Desktop integration for selecting local models and managing app connections. It also fixes cancelled or resumed prefill caching, Claude Code prompt behavior, and DeepSeek Harness startup fallback.

  30. FeatureAug 19· GitHub Releases

    v0.32.15

    Ollama v0.32.15 adds a model metadata cache to reduce per-request overhead.

  31. FeatureAug 15· GitHub Releases

    v0.32.14

    Ollama v0.32.14 adds WebP image transcoding for llama-server and updates Qwen renderers to accept system messages that are not first in the conversation.

  32. FeatureAug 14· GitHub Releases

    v0.32.13

    Ollama v0.32.13 adds support for developer instructions when running Qwen3.8 models.

  33. FeatureAug 14· GitHub Releases

    v0.32.12

    Ollama v0.32.12 adds support for the Qwen 3.8 27B model. The release includes Apple Silicon optimizations aimed at repeated tasks and coding-agent workloads.

  34. FeatureAug 14· GitHub Releases

    v0.32.11

    Ollama 0.32.11 adds launch support for DeepSeek Harness and Meta's Muse Code agentic coding CLI. Its OpenAI-compatible Responses API now supports web search.

  35. FeatureAug 12· GitHub Releases

    v0.32.10

    Ollama 0.32.10 changes the default repeat penalty from 1.1 to 1.0, improving speculative decoding compatibility and speed. It also accelerates NVFP4 MLX prefill and fixes OCI blob verification.

  36. FeatureAug 11· GitHub Releases

    v0.32.9

    Ollama 0.32.9 adds the Nemotron 3 architecture and support for NVIDIA’s open Nemotron 3.5 Lightning MoE model. It also fixes a boundary condition in Muse Glimmer function-call parsing.

  37. NewsAug 11· Changelog

    Ollama changelog updated

    :root { --tab-size-preference: 4; } pre, code { tab-size: var(--tab-size-preference); } Releases · ollama/ollama · GitHub Skip to content Navigation Menu Sign in Appearance settings Platform AI CODE CREATION GitHub Copilot Write better code with AI GitHub Copilot app Direct agents from issue to merge MCP Registry Integ

  38. NewsAug 10· GitHub Releases

    v0.32.8

    ## What's Changed * Add Muse Glimmer support for NVIDIA, AMD, and additional platforms **Full Changelog**: https://github.com/ollama/ollama/compare/v0.32.7...v0.32.8-rc0

  39. NewsAug 10· GitHub Releases

    v0.32.7

    ## Muse Glimmer > Note: Muse Glimmer is currently available via initial support via Ollama's MLX engine on Apple Silicon. Additional support and optimizations for Apple Silicon, NVIDIA, AMD, and other platforms will be available in the coming days. **Muse Glimmer**, Meta's newest open model and the first released

  40. NewsAug 4· GitHub Releases

    v0.32.6

    ## What's Changed - Qwen3.5 is faster on Apple GPUs: the MLX engine now uses the model's MTP head for speculative decoding automatically - `/v1/chat/completions` streaming 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 `str