I can hardly believe my eyes. Intern Science just dropped Agent A1. Absolutely next level performance and q4_km fits in just 21gb of ram or VRAM! Check it out here: https://huggingface.co/InternScience/Agents-A1-Q4_K_M-GGUF

Thanks to the incredible team at DeepReinforce AI. We have a 9b model that's being compared with qwen3.5 35b. AND IT FITS IN AN 8GB GPU/LAPTOP. State-of-the-Art Coding Agents: Available in 9B-Dense, 31B-Dense, 35B-MoE, and 397B-MoE (post-trained on top of Gemma 4 and Qwen 3.5), achieving state-of-the-art performance among open-source models of comparable size on coding benchmarks such as Terminal-Bench 2.1, SWE-Bench, NL2Repo and OpenClaw. Self-Improving Training Framework: Ornith-1.0 employs RL to learn to generate not only solution rollouts, but also the scallfold that drive those rollouts. By jointly optimizing the scaffold and the resulting solution, the model discovers better search trajectories and generates higher-quality solutions. Licence: MIT licensed, globally accessible, and free from regional limitations. I've used this model for 48 hours now and the results are very promising (posting results in a few days!) Check it out here: https://huggingface.co/deepreinforce-ai/Ornith-1.0-9B-GGUF

If you've used image or video gen models then you kind of already know what Diffusion is, without actually knowing what it is. Diffusion is the art of processing your request in patterns rather than linearly/sequentially. When you talk to say GPT 5.5, your prompt/message is being processed word for word, token by token. To ensure GPT can respond properly. Quite an expensive way to process prompts. It's like talking to a PHD level expert and asking them how planes stay in the air, then immediately ask how to make a cheese sandwich. Your expert will answer both questions properly but it will cost the expert it's brain cells and capacity. In plain english this means diffusion models don't need to process each word/token in a sequence. Which saves A LOT on compute power. What's the real upside? Over 1,000 tokens per second. On consumer hardware like @Mason Page 's. Mind boggling speeds right? And we're just getting started. This is the 3rd generation of Diffusion LLMs on the market and the first from Google. Can't wait to see other providers building in the diffusion space soon because damn guys we're going into hyperspace mode!! What could you do with 1,000 tokens per second speeds? How many websites, apps, softwares, or client solutions could you build IN A DAY now? Loads I say.

Live replay edited :)

Qwen3.5 9B has seriously exceeded my expectations.