Recursive's Flint: 40B local coding MoE, quantized to run on a 24GB Mac
_rockt · x · 2026-10-24
Atomic Chat released a GGUF quant ladder (42.9GB down to 13.3GB) for Recursive's first model Flint, recommending the 13.3GB AD-IQ2XXS for 24GB hardware with 95K tokens of context headroom.
Key points:
- Flint is a 40B coding MoE (6 of 256 experts active), built from scratch in under 4 months by Recursive's automated research system, the Eureka Machine
- Claimed 125+ tokens/sec on a MacBook Pro and 150+ on RTX Spark — up to 2x faster than comparable local models
- Open weights on Hugging Face; runnable locally or free via opencode and pidot
- The chart shows Top-1 agreement with BF16: AD-IQ4XS (24.5GB, 91.25%), AD-Q4KM (25.8GB, 91.72%), AD-Q6K (34.7GB, 94.67%)
More from coding & agent
- Should an AI call agent keep running when one tool goes down? A degraded-mode proposal — SubtleInterval_4449 · 2026-10-24
- Claude Desktop blocks local models; users bypass it with a loopback proxy trick — pneuny · 2026-10-24
- OpenAI ships Automations in Agents API for scheduled agent tasks — OpenAIDevs · 2026-10-24
- Arca runs its financial services business on long-horizon agents, not with them — AI Engineer · 2026-10-24
- AGENTS.md explained: one root Markdown file to rule every AI coding tool — adnan_hashmi · 2026-10-24
- Summarizer hallucinates details from headline-only feeds, dev asks how to check faithfulness — rahul_watertech · 2026-10-24