Image created with gemini-2.5-flash-image with claude-sonnet-4-5. Image prompt: Cinematic black and white photograph of dramatic cloudscape with crepuscular rays breaking through fragmenting cumulus clouds, multiple beams of light scattering in different directions across sky, high contrast film photography, bold sans-serif title card reading ‘OPEN SOURCE’ in bottom third, minimal composition focused on sky, film grain texture
The HuggingFace team just got Claude Code to fully train an open LLM. You just say something like: “Fine-tune Qwen3-0.6B on open-r1/codeforces-cots.” Claude handles the rest. ▸ Picks the best cloud GPU based on model size ▸ Loads dataset (or searches if not specified) ▸ https://x.com/LiorOnAI/status/1997754848255807874
We Got Claude to Fine-Tune an Open Source LLM https://huggingface.co/blog/hf-skills-training
After a bit of work (FP8 dequant, some debugging), got Mistral Vibe working on my M3 Max w/ Devstral 2 Small, running locally w/ LMStudio. It worked quite well in a quick demo, as shown below (everything below is happening 100% locally) – video at 4x speed: https://x.com/N8Programs/status/1998591943798882484
Congrats to @MistralAI on the big launch 🚀 You can check out Mistral Vibe in Zed today – just download it and add your API key! https://x.com/zeddotdev/status/1998456122886238589
Devstral 2 beats or ties Deepseek v3.2 71% of the time by third party preference and is smaller/faster/cheaper (esp Small 2) !!! i think there’s a good chance @theo spoke too soon -Mistral is just starting out on a very epic comeback. Nextgen will get >10x compute and i’d bet https://x.com/swyx/status/1998600513538109476
Devstral 2 is #1 trending on @OpenRouterAI! Try it for free in Mistral Vibe: → uv tool install mistral-vibe https://x.com/MistralAI/status/1999124574798217324
Introducing the Devstral 2 coding model family. Two sizes, both open source. Also, meet Mistral Vibe, a native CLI, enabling end-to-end automation. 🧵”” / X https://x.com/MistralAI/status/1998407332502405347
Introducing: Devstral 2 and Mistral Vibe CLI. | Mistral AI https://mistral.ai/news/devstral-2-vibe-cli
Okay this is also sick: the Mistral Vibe CLI bootstraps itself with uv https://x.com/charliermarsh/status/1998447013797458336
Our next-generation coding model family Devstral 2 is available in two sizes: Devstral 2 (123B) under a modified MIT license, and Devstral Small (24B) under Apache 2.0. Both SOTA, open-source, free to use, and available now via our API. https://x.com/MistralAI/status/1998407335308358028
HuggingFace just made fine-tuning 10x easier! One line of English to fine-tune any open-source LLM. They released a new “”skill”” you can plug into Claude or any coding agent. It doesn’t just write training scripts, but actually submits jobs to cloud GPUs, monitors progress, and https://x.com/akshay_pachaar/status/1997946287556321359
Glad to release the new devstral, the best open model for code agents! It comes with vibe CLI, and you can install it easily with uv or pip and let it guide you! uv tool install mistral-vibe pip install mistral-vibe”” / X https://x.com/b_roziere/status/1998408872168391166
Congrats to the @MistralAI team on the launch of Devstral 2! 🚀 vLLM now delivers Day-0 support for the Devstral 2 Instruct models — optimized for agentic coding, deep codebase exploration, and multi-file editing at scale. Feel free to reach out 👇 https://x.com/vllm_project/status/1998428798891765926
Directly comparing a benchmark of Devstral2-123B on my hardware to MiniMax-M2 (230B-A10B) shows the difference in performance MoE can give. At 100 requests concurrently: MiniMax is 2x faster At 2 requests concurrently: MiniMax is 3.5x faster https://x.com/JustinWaugh/status/1998467712235028888
ollama run deepseek-v3.2:cloud DeepSeek v3.2 is now on Ollama’s cloud! DeepSeek v3.2 on Ollama’s cloud can have thinking enabled and disabled. Give it a try. It’s free to get started! https://x.com/ollama/status/1998293403801706613
13 secs for 5 GB between @huggingface & @googlecloud thanks to our new collaboration 🤯🤯🤯 https://x.com/ClementDelangue/status/1998157804020941044
Meta AI’s Saber redefines zero-shot reference-to-video generation It generates stunning, identity-preserving videos from text & images. No costly R2V datasets required, trained solely on video-text pairs. Achieves state-of-the-art with masked training. https://x.com/HuggingPapers/status/1998485543345131847
🚀 Opportunities in the Mistral Warsaw 🧜♀️ office! We’re hiring AI Scientists 🧑🔬 and Research Engineers🆕to join our team. https://x.com/PiotrRMilos/status/1999071560943182124
nice that it’s open weight, but comparing dense vs moe models and only looking at total params is pretty unfair, if you look at active params instead of total params it’s a different story: – GLM 4.6 (32B): 74% fewer – Minimax M2 (10B): 92% fewer – K2 thinking (32B): 74% fewer – https://x.com/eliebakouch/status/1998427299788550450
i should clarify, no hate on mistral here, i’m really glad they release it in open weight and i’m sure it will be useful for researcher! It’s the comparison on total parameters for dense vs moe that i find very misleading but it’s not something that they advertise heavily (it’s”” / X https://x.com/eliebakouch/status/1998436178714882330
Thanks to the Textual library we were able to build a pretty neat UI for Mistral Vibe in the terminal Looks especially good with Ghostty ✨ https://x.com/onetwoval/status/1998439440797020527
Thats a pretty restrictive license that will make using these Mistral models or building on them impossible for most organizations. Even if you don’t make $20M/month a potential acquirer may. Historically, a lot of labor for open source projects comes from for-profit firms.”” / X https://x.com/emollick/status/1998509960347709842
uv tool install mistral-vibe”” / X https://x.com/qtnx_/status/1998407441256530163
Very excited to release two new open-weight models, Devstral 2 (123B) and Devstral Small 2 (24B), along with Mistral Vibe, a CLI built for Devstral that enables end-to-end code automation! https://x.com/GuillaumeLample/status/1998409211068232119
Moondream segmenting delivers pixel-accurate vector masks for arms, packages, and conveyor bins. A free, open-source model with state-of-the-art segmentation benchmarks, built for real-world automation. https://x.com/moondreamai/status/1998465589027967201
DeepSeek is Using Banned Nvidia Chips in Race to Build Next Model — The Information https://www.theinformation.com/articles/deepseek-using-banned-nvidia-chips-race-build-next-model
Today we open source Nomos 1. At just 30B parameters, it scores 87/120 on this year’s Putnam, one of the world’s most prestigious math competitions. This score would rank #2/3988 in 2024 and marks our first step with @hillclimbai towards creating a SOTA AI mathematician. https://x.com/NousResearch/status/1998536543565127968?s=20
Today we’re open-sourcing a preview of our two new models in the Isaac family: hybrid-reasoning 2B and 1B-parameter best-in-class vision-language models. Weights → https://x.com/perceptroninc/status/1998812935821697363
Holy moly, thats insane: Nomos 1 is a 30B open-source model that just scored 87/120 on this year’s Putnam, good enough for an estimated #2/3988, showing that near-top human math performance is now possible with relatively small models plus good post-training and reasoning https://x.com/kimmonismus/status/1998749650984255985
🚀 Qwen3-Omni-Flash just got a massive upgrade (2025-12-01 version) ! What’s improved: 🎙️ Enhanced multi-turn video/audio understanding – conversations flow naturally ✨ Customize your AI’s personality through system prompts (think roleplay scenarios!) 🗣️ Smarter language https://x.com/Alibaba_Qwen/status/1998776328586477672
NousResearch comes in HOT 🚨 > specialised Qwen3-30B-A3B-Thinking-2507 > math solving > proof writing https://x.com/ariG23498/status/1998654584529797522
Releasing jina-VLM: our new 2B vision language model achieves SOTA on multilingual visual question answering and document understanding among open 2B-scale VLMs. https://x.com/JinaAI_/status/1997926488843190481
Leave a Reply