Last updated · 70 releases tracked since August 2023
A dated log of everything Ollama has shipped, collected from official Ollama sources and updated as new releases land.
Ollama released v0.35.0 and v0.35.1 in September 2026, introducing decision models via /v1/systemone and support for Cloudflare's Clef and Jev-style models. Key updates included ChatGPT Desktop integration, MLX memory improvements, and enhanced structured outputs.
Summary of September 2026, generated from the releases listed below.
This month
0
-9 vs last month
Last 12 months
42
Busiest: August 2026
Most shipped
Developer Tools
19 releases
Newest first. Each entry links to the official page it came from.
Ollama v0.35.1 adds support for Cloudflare's Clef (27B) and Clef Flash (9B) decision models, enabling multimodal input with images and text. Enhancements include up to 10 web searches per response and modelfile capability declarations.
Ollama now supports Jev-style decision models via TypeSafe's API, enabling fast, low-cost yes/no decisions with models like Nimble 9B. Use cases include ticket triage and content moderation.
Ollama v0.35.0 introduces decision models via /v1/systemone, enabling tasks like ticket triage with models Nimble and Tev1. Supports choice, noul, and score question types.
v0.34.4 includes faster structured outputs, fixed 'model not found' errors, macOS stability, Qwen 3.8 performance improvements, Gemma 4 image resolution enhancements, and updated libraries.
New thinking controls for models, Apple Silicon support for Nemotron H, macOS app stability fix, and HuggingFace pull improvement
v0.34.2 includes first-run setup with sign-in/local options, ollama://apps link, memory growth fix with MLX speculative decoding, and llama.cpp update.
MLX safetensors no longer experimental; GGUF model creation now uses llama.cpp tooling. Improved MLX memory handling on Apple Silicon, enhanced repeat token detection, faster /api/tags endpoint, and deprecation of 'typical_p'
Integrates Ollama models into ChatGPT Desktop for macOS, improving structured output performance on Apple Silicon and adding OpenAI-compatible client tool search and response compaction.
Ollama v0.33.3 adds gemma4 support for images/audio on MLX, cached prompt tokens, and updates MLX/llama.cpp engines with improved parameter handling
Ollama introduces transparent per-token pricing for Pro, Max, and Team plans, including monthly usage credits. Free plan now offers starter models with pay-as-you-go options. No hidden fees, zero data retention, and dedicated US/Europe compute.
Ollama v0.33.2 includes dark mode restoration, macOS app improvements, and Claude Desktop proxy fixes.
Ollama v0.33.1: MLX Qwen3.8 Flash Next support, cmake updates, structured output in mlxrunner, and GPU timeout fixes. New contributor @pd95.
Claude Desktop can now use Ollama as a third-party gateway to access open models, both local and cloud-based, with privacy protections and seamless switching between models.
v0.33.0 release includes Claude Desktop integration, improved caching with prefill reliability, and various bug fixes.
v0.32.15 includes a new desktop onboarding flow, improved TTFT via metadata caching, bug fixes, Qwen 3.8 system message normalization, and dependency updates
Updated LLM to transcode WebP images for llama-server and improved Qwen renderers to handle non-leading system messages.
Support for developer instructions in qwen3.8.
Adds Qwen 3.8 27B model with enhanced coding, research, and agentic tasks support. Optimized for Apple Silicon with mlx version.
Added DeepSeek Harness support, Muse Code integration, web search in OpenAI API, and Glimmer template updates
Introduces NVIDIA Nemotron 3.5 Lightning, a 30B MoE model with 3B active parameters for always-on agents, plus parser improvements.
NVIDIA Nemotron 3.5 Lightning, a 30B parameter open model with 3B active parameters, now available on Ollama for local agentic tasks, tool calling, and multi-step workflows.
Muse Glimmer model support expanded to NVIDIA, AMD, and additional platforms with enhanced performance on Apple Silicon via MLX engine.
Meta's 30B multimodal Muse Glimmer model now available via Ollama with MLX engine support for coding agents and image processing.
Ollama raises $88M to advance open models, enabling developers to run and customize AI locally with ownership, affordability, and privacy.
Ollama 0.31 enhances Gemma 4 performance on Apple Silicon by 90% via MLX-powered multi-token prediction, boosting coding agent responsiveness without altering output.
Ollama's MLX engine achieves highest Apple Silicon performance with NVFP4 support, 20% faster output, and improved agent workflows.
Ollama 0.30 enhances performance with GGUF compatibility, Vulkan support, and 20% faster NVIDIA GPU throughput, expanding model and hardware compatibility
NVIDIA Nemotron 3 Ultra: 550B parameter open model for long-running agents, 1M token context, and high-efficiency reasoning. Available on Ollama's cloud with cost-efficient performance.
OpenJarvis v1.0 released: open-source framework for local AI agents with built-in Ollama support
Ollama now leverages MLX framework for improved performance on Apple Silicon, including NVFP4 support, caching improvements, and faster inference speeds.
Setup OpenClaw in 2 mins with a single Ollama command. Features include web search, messaging app integration, and local/cloud model support. Easy installation and configuration.
Ollama added subagents and built-in web search for Claude Code without needing MCP servers or API keys, improving agentic workflows in the shell.
Ollama introduced OpenClaw support, a local-first assistant gateway that connects messaging apps to coding agents and can be launched directly with ollama launch openclaw.
Ollama introduced `ollama launch`, a one-command setup for Claude Code, OpenCode, Codex, and other coding tools.
Ollama added experimental image generation on macOS with Z-Image Turbo, with Windows and Linux support coming soon.
Ollama made v0.14.0+ compatible with the Anthropic Messages API, so Claude Code can run against local or cloud Ollama models.
Open models can be used with OpenAI's Codex CLI through Ollama. Codex can read, modify, and execute code in your working directory using models such as gpt-oss:20b, gpt-oss:120b, or other open-weight alternatives.
Ollama partners with OpenAI and ROOST to release gpt-oss-safeguard models (20B/120B) for safety classification, Apache 2.0 licensed.
MiniMax M2, a coding and agentic workflow model, now on Ollama's cloud. Features superior intelligence, advanced coding, agent performance, and efficient design. Integrates with VS Code, Zed, Droid, and API.
Ollama releases performance benchmarks for NVIDIA DGX Spark, testing models like gpt-oss, gemma3, and llama3.1 with metrics on prefill and decode speeds.
Ollama introduces GLM-4.6 and Qwen3-coder-480B with enhanced tool integration and faster Qwen3-Coder-30B for cloud and local use.
Ollama now supports Alibaba's Qwen3-VL, a vision language model with advanced capabilities in visual agent, coding, spatial perception, and multimodal reasoning. Available via CLI, API, and libraries for text and image processing.
Ollama introduces a web search API with a free tier and cloud-based rate limits, enhancing model accuracy by reducing hallucinations. Integrates with Python and JavaScript libraries for research tasks.
Ollama introduces improved model scheduling system with exact memory management, reducing OOM crashes, maximizing GPU utilization, and enhancing multi-GPU performance. Includes metrics improvements and support for new models.
Ollama Cloud models now in preview, enabling larger models on datacenter-grade hardware with privacy and seamless local/cloud integration. Supports OpenAI API and CLI tools.
Ollama's new macOS/Windows app enables model chatting, file drag-and-drop, multimodal support, and documentation processing with adjustable context length.
Ollama introduces Secure Minions, a secure protocol enabling encrypted local-remote collaboration with frontier models. Encrypts context, reduces cloud costs 5x-30x, and ensures privacy with NVIDIA Hopper H100 GPUs.
Ollama introduces 'Thinking' feature, allowing users to enable/disable model's thinking process for tailored application behavior. Supports CLI, API, and libraries with options to separate thinking from output or bypass it for faster responses.
Ollama now supports streaming responses with tool calling, enabling real-time chat applications to stream content and call tools simultaneously.
Ollama introduces a new engine for multimodal models, supporting Llama 4, Gemma 3, Qwen 2.5 VL, and Mistral Small 3.1 with improved accuracy, memory management, and modularity.
Ollama collaborates with cloud models (e.g., GPT-4o) to shift workloads to local devices, achieving 30.4x cost reduction with 87% performance. MinionS protocol reduces costs 5.7x while maintaining 97.9% performance. Open-source code available.
Ollama introduces structured outputs, enabling JSON schema-defined responses. Updates to Python/JavaScript libraries for data parsing, image extraction, and consistent model outputs.
Ollama Python library 0.4 introduces function calling improvements, full typing support, and new examples for better integration and development.
Llama 3.2 Vision 11B and 90B models now available in Ollama with VRAM requirements (8GB/64GB) and CLI/library integration options.
Ollama partners with IBM to bring Granite 3.0 models under Apache 2.0 license, including dense and MoE variants with RAG support and performance comparable to state-of-the-art open models.
Ollama partners with Meta to bring Llama 3.2, offering small (1B, 3B) and multimodal (11B, 90B) models. Local processing ensures privacy, with text-only and vision capabilities.
Ollama introduces tool calling support for models like Llama 3.1, enabling complex tasks via functions, APIs, and web browsing. Includes OpenAI compatibility and API examples.
Google Gemma 2 is now available on Ollama in 3 sizes - 2B, 9B and 27B. Features a new architecture with class-leading performance and efficiency, surpassing models twice its size in benchmarks.
Ollama's Continue enables open-source AI code assistance in VS Code and JetBrains using models like Codestral, DeepSeek Coder, and Llama 3, with local deployment options
Google announces Firebase Genkit with Ollama support for running Google’s open-source Gemma model locally.
Llama 3, Ollama's next-gen LLM, offers 7x larger dataset, 8K context, 128K tokens, and 8B/70B variants. Improved efficiency and reduced refusals over Llama 2.
Ollama introduces embedding models for generating vector embeddings in RAG applications, supporting models like mxbai-embed-large and integrations with LangChain and LlamaIndex.
Ollama now supports AMD graphics cards in preview on Windows and Linux, accelerating all features. Supported cards include Radeon RX 7900 XTX, 7900 XT, 7900 GRE, and others. More AMD support coming soon.
Ollama now available on Windows in preview with GPU acceleration, full model library access, and OpenAI-compatible API. Runs models natively, supports vision models, and includes always-on API server.
Ollama adds initial compatibility with OpenAI Chat Completions API for local model integration
New LLaVA 1.6 vision models available in 7B, 13B, and 34B sizes with enhanced image resolution, text recognition, and reasoning capabilities.
Ollama releases Python and JavaScript libraries for easy integration, offering full REST API features, compatibility with Ollama versions, and support for streaming, multi-modal, and text completion. Available on GitHub.
Ollama now available as official Docker image with GPU acceleration support for Linux and CLI/API access. Runs locally without third-party data sharing.
Meta's Code Llama is now available on Ollama for local execution with 7B, 13B, and 34B parameter options. Commands include 'ollama run codellama:7b' and specialized Python versions.
Ollama enables running uncensored Llama 2 models locally, offering comparisons with censored versions. Includes examples like llama2-uncensored and nous-hermes-llama2 for enhanced flexibility.
Releases per month over the last 12 months. Busiest month: August 2026, 14 releases.
Releases in the last 12 months by category and product.
The most recent Ollama update we track is “Ollama v0.35.1”, released on September 29, 2026.
Ollama shipped 9 updates in September 2026, including “Ollama v0.35.1”, “Ollama now supports Jev-style decision models”, “Ollama v0.35.0” and 6 mores.
We tracked 42 updates from Ollama in the last 12 months, about 4 a month. The busiest month was August 2026, with 14 releases.
Newsletter
This viewscreen turns the calendar into a recurring digest: the sharpest launches, model drops, and product moves delivered in one clear read.
The standout launches, model drops, and platform moves worth tracking.
One tight read instead of bouncing between company blogs and timelines.
Clear summaries you can scan fast and forward to the team when needed.
Monthly dispatch
Get a clean monthly dispatch covering the launches that mattered, not every minor product breadcrumb.
One email each month, tuned for signal over noise.
