What Tablif read on Flint's site
Flint describes itself as: “Introducing Flint - Recursive”
- Stage: Live, no waitlist
- How it differs: “Flint leads MoE models with fewer than 50 billion total parameters on agentic coding benchmarks”
What it is
Calls itself “mixture-of-experts (MoE) model”.
- Calls itself: mixture-of-experts (MoE) model
“…mixture-of-experts (MoE) model”
Where it puts itself
Claims “best-performing”.
- Superlative claim: best-performing
“Flint leads MoE models with fewer than 50 billion total parameters on agentic coding benchmarks”
“Flint is the best-performing MoE model with fewer than 50 billion total parameters”
What it costs
No free plan.
- Free plan: no
“Flint’s weights are available on Hugging Face under the Apache 2.0 license.”
Who it sells to
Sells to developers.
- Sells to: developers
“Use Flint as your daily coding model for tasks you want to run locally”
How it runs
A desktop app for Apple Silicon and NVIDIA. Runs locally and works offline. Distributed through the Hugging Face. AI is the core of the product. Built with Framer.
- Kind: desktop
- Runs on: Apple Silicon · NVIDIA
- Runs where: on the device, offline
- Distributed via: Hugging Face
- AI role: core to the product
- Built with: Framer
“Download the model, run it locally”
“…on Apple Silicon and NVIDIA RTX Spark”
What it works with
Names Eureka Machine, LM Studio and 5 more on its pages.
- Names: Eureka Machine · LM Studio · Ollama · +4
“Building Flint was a practical test of Recursive’s Eureka Machine, our system for large-scale automated AI research and development.”
“Introducing Flint, a fast, intelligent model for coding locally on your laptop”
What it offers as proof
Live, no waitlist. Funding: “$4 Billion”. Publishes docs and a careers page.
- Funding: $4 Billion
“Flint’s weights are available on Hugging Face under the Apache 2.0 license. Download the model, run it locally with your data kept on your device, and adapt it for your own applications.”
“Notable Researchers Join $4 Billion Effort to Build Self-Improving A.I.”
Sources
Tablif read these pages, last on 23 Oct 2026. Everything above is what the product says about itself; Tablif does not verify every claim.
Similar products in Tablif Town
- Agentic Workbench: Supervised coding agents with reviewable changes
- Pipeline Simple: Make more money with GEO(Generative Engine Optimization)
- AgentTaste: Production-tested SKILL.md packages for agent
- The Agentic Coding Kit: Guardrails and review playbooks for AI coding assistants
- DevGems: Curated AI coding tools for developers who ship faster
- Reqode: Give AI agents the context to build whole features.
- Muse Wonder: Meta's frontier model for personal and professional agents
- Agent Maestro: Autonomous software development with AI coding agents