Introducing Flint, @Recursive_SI’s first model: a fast local coding...
Richard Socher@RichardSocher
10 views
Oct 22, 2026
~2 min read
Advertisement
1
Introducing Flint, @Recursive_SI’s first model: a fast local coding model built with our automated research system, the Eureka Machine.
• Best-in-class fast local coding
• Up to 2x faster than comparable models: 125+ tokens/sec on a MacBook Pro, 150+ on RTX Spark
• Built from scratch in under 4 months
Private, unmetered intelligence on your own machine. Open weights available now on @huggingface, run it locally on your favorite framework or try it for free on @opencode or @pidotdev.
• Best-in-class fast local coding
• Up to 2x faster than comparable models: 125+ tokens/sec on a MacBook Pro, 150+ on RTX Spark
• Built from scratch in under 4 months
Private, unmetered intelligence on your own machine. Open weights available now on @huggingface, run it locally on your favorite framework or try it for free on @opencode or @pidotdev.
2:41
2
Flint pushes the Pareto frontier on local coding.
3
Flint is the best-performing model in its class (MoE models under 50B) on agentic coding benchmarks like SWE-bench Pro v2, Terminal-Bench 2.1, and SWE-bench Verified.
It also rivals similarly sized dense models that are much slower to run.
It also rivals similarly sized dense models that are much slower to run.
4
Flint is also the fastest model in its class:
• 125+ tokens/sec on an M4 Pro MacBook Pro
• 150+ tokens/sec on a 64GB NVIDIA RTX Spark
Up to 2x faster than comparable models on short contexts, and 5x+ faster than dense models with similar coding scores. Your data stays on your computer, and it works offline.
• 125+ tokens/sec on an M4 Pro MacBook Pro
• 150+ tokens/sec on a 64GB NVIDIA RTX Spark
Up to 2x faster than comparable models on short contexts, and 5x+ faster than dense models with similar coding scores. Your data stays on your computer, and it works offline.
5
Flint is also a real-world test of the Eureka Machine, our system for automated AI research.
It runs the full research loop: ideation, implementation, and validation. Then it uses the results to decide what to try next. We used it in nearly every part of how Flint was built.
This is what allowed our very small team to build this model in <4 months.
Its contributions include a sparser model shape, low-bit training that shrank Flint by 71% with almost no loss in quality and custom kernels.
It runs the full research loop: ideation, implementation, and validation. Then it uses the results to decide what to try next. We used it in nearly every part of how Flint was built.
This is what allowed our very small team to build this model in <4 months.
Its contributions include a sparser model shape, low-bit training that shrank Flint by 71% with almost no loss in quality and custom kernels.
6
Try Flint today.
Read the blog: recursive.com/updates/introd…
Download on Hugging Face: huggingface.co/RecursiveSI/fl…
Run Flint locally with @ollama, @lmstudio, @UnslothAI, @exolabs, @Nativ_AI, @sgl_project, @vllm_project and llama.cpp or try it for free on @opencode and @pidotdev
Read the blog: recursive.com/updates/introd…
Download on Hugging Face: huggingface.co/RecursiveSI/fl…
Run Flint locally with @ollama, @lmstudio, @UnslothAI, @exolabs, @Nativ_AI, @sgl_project, @vllm_project and llama.cpp or try it for free on @opencode and @pidotdev