Ollama v0.40.8 adds Recursive's Flint coding model with 262K context
modelOllama v0.40.8 ships support for Flint, Recursive's Apache 2.0-licensed 40B MoE coding model (~2.5B active params) built for repo-scale agent coding: understanding unfamiliar code, tracing bugs, implementing features, and writing tests. Features reasoning, multi-turn tool use, and a 262K-token context window. Reported 54.9% on Terminal-Bench 2.1 and 74.2% on SWE-bench Verified. Mac-only for now.
An agent can now run Recursive's Flint — a 40B MoE coding model with 262K context, reasoning, and multi-turn tool use — locally via `ollama run flint` for repo-wide coding tasks.
Model release wrapped in an Ollama runtime update. Flint is positioned specifically for coding agents (repo-scale understanding, tool use, long context). macOS-only at launch — note this limits deployment surface. SWE-bench Verified 74.2% is competitive if reproducible.
ollama run flint# Install/upgrade Ollama to v0.40.8, then pull & run Flint
ollama run flint
# Or call via HTTP API for agent use
curl http://localhost:11434/api/chat -d '{
"model": "flint",
"messages": [
{"role": "user", "content": "Trace the bug in src/auth/session.py and propose a fix."}
],
"stream": false
}'