Nick - AI Spotlight News

Showcasing the latest AI Updates/News 🗞️
Breakthroughs & Model Comparisons 🚀
Stay Ahead in the AI/Crypto game! 🧠
AI Stock/Web3 News 📊

[Nick - AI Spotlight News @AiSpotlightNews]

Interested in the Web3 or AI community? Comment below to connect with other Web3/AI users and join the journey!

[Nick - AI Spotlight News @AiSpotlightNews]

@WhaleInsider Crypto going down again here we go

[Nick - AI Spotlight News @AiSpotlightNews]

@Rashyv9 Grab

[Nick - AI Spotlight News @AiSpotlightNews]

@elonmusk @bot Everyone should

[Nick - AI Spotlight News @AiSpotlightNews]

@bot @elonmusk Awesome!

[Nick - AI Spotlight News @AiSpotlightNews]

@elonmusk @bot Grok @Bot is awesome!

[Nick - AI Spotlight News @AiSpotlightNews]

@LunaX_Web3 Yes

[Nick - AI Spotlight News @AiSpotlightNews]

@maijiazhangjun yes

[Nick - AI Spotlight News @AiSpotlightNews]

@xxlenaxx0 GM everyone

[Nick - AI Spotlight News @AiSpotlightNews]

@maijiazhangjun on it

[Nick - AI Spotlight News @AiSpotlightNews]

@ItzMeAddie yes

[Nick - AI Spotlight News @AiSpotlightNews]

@Lunayuczt let's connect

[Nick - AI Spotlight News @AiSpotlightNews]

@Nylorien gm

[Nick - AI Spotlight News @AiSpotlightNews]

@hansonhihihi real

[Nick - AI Spotlight News @AiSpotlightNews]

@Shen_ke2 gm

[Nick - AI Spotlight News @AiSpotlightNews]

@Lakhiislam112 hi

[Nick - AI Spotlight News @AiSpotlightNews]

@GraceOfGodH hi

[Nick - AI Spotlight News @AiSpotlightNews]

Recursive just dropped Flint, an open-source coding model built to run on your laptop. 💻

It's the first model from Richard Socher's startup. It has 40B parameters, but only about 2.5B are active per token, so it runs fast on local hardware. The weights are open under Apache 2.0.

What stands out:

Speed: 125+ tokens/sec on an M4 Pro MacBook Pro and 150+ on a 64GB RTX Spark ⚡

Benchmarks: 89.7% on SWE-bench Pro v2, 74.2% on SWE-bench Verified and 54.9% on Terminal-Bench 2.1 📊

Class leader: Recursive says it beats every MoE model under 50B on agentic coding, and that it matches dense 25–40B models while running 5x+ faster 🏆

Built by AI: Recursive's "Eureka Machine" automated research system helped build nearly every part of the model stack, from scratch in under 4 months 🤖

Runs anywhere: works with Ollama, LM Studio, llama.cpp, vLLM and SGLang 🔧

Serious coding help, fully offline, with no API bill.

[Nick - AI Spotlight News @AiSpotlightNews]

Source: https://t.co/hAAKM8zc4l

[Nick - AI Spotlight News @AiSpotlightNews]

Recursive just dropped Flint, an open-source coding model built to run on your laptop. 💻

It's the first model from Richard Socher's startup. It has 40B parameters, but only about 2.5B are active per token, so it runs fast on local hardware. The weights are open under Apache 2.0.

What stands out:

Speed: 125+ tokens/sec on an M4 Pro MacBook Pro and 150+ on a 64GB RTX Spark ⚡

Benchmarks: 89.7% on SWE-bench Pro v2, 74.2% on SWE-bench Verified and 54.9% on Terminal-Bench 2.1 📊

Class leader: Recursive says it beats every MoE model under 50B on agentic coding, and that it matches dense 25–40B models while running 5x+ faster 🏆

Built by AI: Recursive's "Eureka Machine" automated research system helped build nearly every part of the model stack, from scratch in under 4 months 🤖

Runs anywhere: works with Ollama, LM Studio, llama.cpp, vLLM and SGLang 🔧

Serious coding help, fully offline, with no API bill.

[Nick - AI Spotlight News @AiSpotlightNews]

Why it matters: Coding agents use up tokens fast, and most of them run on paid cloud APIs. A model this small and fast makes a free coding agent that runs on your own high-end laptop realistic.

The catch: on a Mac, Recursive recommends 48GB+ of unified memory. The scores are self-reported, and SWE-bench Pro v2 is Recursive's own version of the benchmark, so it doesn't compare directly with public leaderboards.