Based on this YouTube video's total view count of 59.2K views and industry-standard rates, the estimated total earning is $41 - $119 through ad revenue. Historical data is not yet available to calculate daily, weekly, or monthly averages.

Explore Is Flint the Killer of Qwen 3.8 27B? (First RSI Model) with 59,270 views, 853 likes, and 128 comments. Experience the impact of this video content that has captured audience attention.

Recursive Flint writes 131 tokens per second on an M4 MacBook Pro, ten times the 13 tok/s of Qwen 3.8 27B. So is Flint the Qwen 3.8 27B killer, and is it really the first RSI model? Per token, yes. Per finished task, not yet. Flint is Recursive AI’s first model, a 39.8B mixture of experts (MoE) coding model with about 2.5B parameters active per token. It’s open weights under Apache 2.0, with a 262K context, text only. Richard Socher’s lab says its Eureka Machine, an automated AI research system, helped build nearly every part of it in under four months. That is why people are calling it the first RSI (recursive self-improvement) coding model. In this video I test the “fastest local coding model” claim against Qwen 3.8 27B, the dense model most of us run for local AI coding today: ▸ MoE vs dense: why Flint really is about 10x faster per token on a 273 GB/s M4 Pro, and why Qwen 3.8 27B still wins 4 of the 6 benchmarks Recursive published ▸ Tokens per second vs minutes per task: Flint’s reasoning is always on, with no effort dial. In one test it used 45,082 tokens to Qwen 3.6 35B’s 2,732 on the same 14 tasks, and finished 8x later on the clock ▸ The 131 tok/s is measured at 1K context. AI coding agents start at 17K to 26K tokens, and one 115K-context run fell to 9 tok/s ▸ The “2x faster” claim: on the same engine (llama.cpp vs llama.cpp), the edge is about 1.34x ▸ Where Flint wins: short, scoped coding agent tasks (12 of 18 runs in 375 s against 1,749 s for Ornith 1.5) ▸ Who should switch: 48GB M4 Pro and M5 Pro MacBook owners, 16GB GPUs with experts offloaded, AMD (use the Vulkan build), and NVFP4 vs GGUF quality (80 on 50 SWE-bench Verified tasks on an RTX PRO 6000 in vLLM) How to run Flint locally: Recursive’s rec engine (Mac only), Ollama, LM Studio, llama.cpp or vLLM. It’s also free on OpenRouter and in OpenCode’s free tier. CHAPTERS 00:00 131 tok/s vs 13: the Qwen 3.8 27B killer? 02:48 Recursive, Richard Socher and the Eureka Machine 04:07 Flint specs: 40B MoE, 2.5B active 05:47 MoE vs dense: why the speed is real 08:03 Tokens per second vs minutes per task 10:13 45,082 tokens vs 2,732 12:34 Qwen has an effort dial. Flint doesn’t 14:11 Where 131 tok/s lives: 1K context 16:47 The 2x claim and the rec engine 18:36 Where Flint actually wins 21:00 Who should switch: Mac, 16GB GPU, AMD, NVFP4 24:13 How to test it on your own machine 26:00 Verdict: is Flint the Qwen killer? SOURCES Recursive launch post and charts: https://www.recursive.com/updates/introducing-recursive-flint Flint model card: https://huggingface.co/RecursiveSI/flint Always-on reasoning (generation settings): https://docs.recursive.com/guides/generation-settings rec engine requirements: https://docs.recursive.com/edge/rec Troubleshooting (looping): https://docs.recursive.com/resources/troubleshooting Unsloth GGUFs: https://huggingface.co/unsloth/flint-GGUF Ollama: https://ollama.com/library/flint OpenRouter (free): https://openrouter.ai/recursive/flint-v1.0:free TechCrunch on the launch: https://techcrunch.com/2026/10/21/recursive-launches-flint-a-laptop-coding-model-it-says-was-largely-built-by-ai/ Dolmario Local AI, Strix Halo test: https://www.youtube.com/watch?v=7QTHTEBrVTA AICodeKing, M4 Max context test: https://www.youtube.com/watch?v=oJ0gSua-fE0 AtomicAgent, Flint vs Ornith 1.5: https://atomicagent.io/blog/flint-vs-ornith-1-5/ NVFP4 in vLLM on an RTX PRO 6000: https://www.reddit.com/r/LocalLLaMA/comments/1xc6j8t/ Benchmarks are company-reported unless a tester is named. #RecursiveFlint #Qwen #LocalLLM

With SocialCounts.org’s view counter, track your YouTube video’s live view count and YouTube likes count in real time with fast, reliable updates.

Watch every YouTube video live view count rise with our real-time YouTube views tracker—built for accuracy and minimal delay.

Follow YouTube real time views as they happen, using our dedicated view counter for YouTube videos.

Get up-to-date live view count on YouTube and see real-time growth with SocialCounts.org’s smart tracking tools.