Ollama 0.40.8 adds Flint, Recursive's model for coding agents, on Macs only for now

Ollama 0.40.8 adds one model, Flint, which brings Recursive's coding model to Ollama — and for now it runs only on a Mac: the release note says Flint is supported on macOS and that support for other GPUs will come in a later release. The release page carries the date 21 October, 16:20 ; the installers it offers were uploaded on 21 October at 21:28 UTC. Its changelog line runs from v0.40.7 to v0.40.8, and Flint is the whole of what the page lists between them: no fixes, no other changes.

What changed

A model built for coding agents.A coding agent is a tool such as Claude Code or OpenCode that does not just answer a question but reads your files, runs commands and edits code over many steps. Flint is designed for the work such agents do across a repository: understanding unfamiliar code, tracing bugs, implementing features and adding tests. Reasoning, multi-turn tool use and a 262K-token context window support longer tasks across files and tool interactions. The context window is how much text the model can hold in view at once — your instructions, the files it has read and the conversation so far. The download table on Ollama's library page writes the same window as 256K ; that is the same window counted in a different unit, not a smaller one.

Big on disk, small per word.Flint is an Apache 2.0-licensed, 40B-parameter mixture-of-experts model with approximately 2.5B parameters active per token. The library page gives the exact count: 39.8 billion total parameters, approximately 2.5 billion active per token. A mixture-of-experts model is split into many smaller "experts", and for each token — a word or a piece of one — only a few of them do the work. So each token costs roughly the computing of a small model, but all of the experts still have to sit in memory, which is why the download is 23GB. Apache 2.0 is a permissive licence: you may use the model in your own work, including commercial work.

The scores, as the publisher reports them.Recursive reports 54.9% on Terminal-Bench 2.1 , and the library page adds 89.7% on SWE-bench Pro v2. Both are coding benchmarks: Terminal-Bench measures tasks done in a command line, the SWE-bench family measures fixing real issues in real repositories. The library page itself says these are publisher-reported results and that evaluation setups and runtimes can affect scores. Read them as the maker's claim, not as a measurement of what you will see.

How to start it. The release gives ollama run flint; the library page addsollama launch claude --model flint for Claude Code and ollama launch opencode --model flintfor OpenCode.

What it means for you

If you have a Mac,this is a local coding-agent model you can try today. Before you pull it, check that the 23GB file fits in your Mac's unified memory — the single pool of memory a Mac's processor and graphics cores share — with room left for macOS, your other apps and the conversation itself. Our guideMac or PC for Running Models Locally explains how to judge that fit and how memory bandwidth sets the speed. If it fits, start it inside the agent you already use with ollama launch claude --model flint or ollama launch opencode --model flint, or on its own withollama run flint.

If you have a Windows or Linux PC with an NVIDIA or AMD card, wait.The release supports Flint on macOS only; support for other GPUs is promised for a later release. Don't download 23GB expecting your graphics card to run it at full speed.

If you judge models by benchmarks,remember that every figure above comes from the maker. The fair test is your own: give Flint a task from your repository that you have already solved, and compare its work with yours.

If you don't use Flint,this release changes nothing else for you. There is no fix in it to wait for, so you can update whenever is convenient.

How to update.The release's list of files carries an installer for each system, all uploaded on 21 October. On Windows, download and runOllamaSetup.exe; on a Mac,Ollama.dmg; on Linux, theinstall.shscript from the same list. Then open a terminal and typeollama -v: it prints the version you have installed, which should now be 0.40.8. If it shows an older number, quit Ollama completely, start it again and run it once more.

The source

The release page on GitHub, Ollama v0.40.8, its list of downloadable files with their upload times, and the model's page in Ollama's library, flint.