simular left the model and the harness alone and only evolved the interfaces, and the same open 27B model now beats GPT-5.5 and Opus 4.8 at desktop tasks for ~$1.10 a task. the interface was the whole fight, nice work team
Our agent scientist, Agent S4, just got an open 27B model to beat GPT-5.5 and Claude Opus 4.8 on OSWorld V2, with the model's weights left exactly as they were.
With the interfaces S4 evolved, Qwen3.8-27B completes 21.3% of the 108 tasks, the highest binary success of any open-source computer-use agent and 3.8x the same model with fixed interfaces.
It costs $1.10 per task, against $25.50 for GPT-5.5 and $72.40 for Opus 4.8 💸
Full write-up: simular.ai/articles/agent-s4…
scenario put 1,300+ ai-made 3d models on hugging face, free, each one with the input image and settings that made it. running the same pic through all of them tonight to see who keeps the fingers
We just open-sourced 1,300+ 3D models.
That's 39 GB of 3D files, free to download on Hugging Face.
huggingface.co/datasets/scen…
> Hitem3D, Meshy, Rodin, Tripo, Claude Opus 5.5 3D and GPT-6 Astra 3D, all run on Scenario
> every example ships its input image, prompt and exact settings
> GLB up to 8K textures, native Blender files from Opus and Astra, CC BY 4.0
IREN is putting NVIDIA's new Minerva on its cloud: software that watches a whole building of GPUs and flags the one about to die before it takes your training run down with it. babysitting that many cards by hand is not a job i want
IREN is partnering with NVIDIA to launch Minerva, NVIDIA’s new AIOps platform, on IREN AI Cloud.
Our VP of AI Cloud, Jason Forrester, joins NVIDIA’s Gilad Shainer and 650 Group’s Alan Weckel to explore how IREN’s approach to AIOps enhances visibility, strengthens operations and keeps compute productive.
Watch the conversation: youtu.be/MWVjiWYs1sw
Abliteration is now renting out DeepSeek's V4.1 Flash with the refusals taken out: images in, a million tokens of context, priced under their own bigger model. selling "cyber-capable" models to defenders by the token, that's a whole storefront, ok
Now live: abliterated-model-large-flash (our version of DeepSeek-V4.1-Flash-1017). It is a cheaper alternative to our GLM-5.3-based model for product owners and defenders who need access to the most cyber-capable models.
- 1M context
- Vision
- Zero input/output retention or training
We believe paying customers should be in control, and over the past few months, leading experts and major brands have adopted our models.
abliteration.ai
nvidia fit a 4b model into a tiny robot brain so reachy can chat and look around your living room without a single frame leaving the house. that's exactly why local wins on a robot, congrats team
Available now: NVIDIA Nemotron 3.5 Bolt E4B 4GB ⚡️
Our newest small open model is built for local assistants, interactive characters and robots that can carry on a conversation and use tools to get things done.
Watch Reachy Mini chat and move, with Bolt handling conversation and NVIDIA Cosmos handling vision. Both models run locally on Jetson Orin Nano 8GB.
Learn more: research.nvidia.com/labs/nem…
tencent just open-sourced a video world model you can walk around in: type a prompt or press WASD and the same world keeps going for a minute instead of resetting every clip. weights, training code, apache. squeezes onto one 4090. congrats team
🚀 Introducing Hy-Nomas: An Open Source Free-Form Interactive Video World Model
Key Features:
* Free-form interaction with text prompts and camera trajectories
* Long-horizon generation with world memory
* Lightweight LoRA adaptation for specialized domains such as robotics
* Optimized for Inference on Consumer GPUs
Test it and drop feedback anytime！
🔗 Website: hy-nomas.hy.tencent.com/
💻 GitHub:github.com/Tencent-Hunyuan/H…
🤗 Hugging Face: huggingface.co/tencent/Hy-No…
🎥 Demo: hy-nomas.hy.tencent.com/demo
#Hy#WorldModels #VideoGeneration #GenerativeAI
jack's team at block trained a voice model on nothing but recordings of people talking, no text at all. it hears you while it's talking and it runs on a laptop. i want to hear it argue with me
We believe voice is the most natural way to interact with intelligence, so we created Dessa, the most authentic full-duplex speech model ever built. Talking with it feels like talking with a person…full of emotion, nuance, pauses, and a natural cadence.
block.xyz/dessa
minimax just put the m3.1 flash weights up, so space bunny is yours to download and host instead of renting it by the token. 428B total, only 23B switching on per token, which is the part that keeps it cheap. nice work team
MiniMax M3.1 Flash weights are now public. 428B parameters total, 23B activated per token.
Agentic performance: A generational leap in agentic tasks, coding, and cybersecurity — 76.2 on BankerToolBench, 73.7% on DeepSWE v1.1, 61.7 on OSWorld V2, 29.6 on Agents' Last Exam, and 82.2 on CyberGym.
Efficient by design: MiniMax Sparse Attention and DSpark speculative decoding improve inference efficiency.
Hugging Face:
huggingface.co/MiniMaxAI/Min…
ModelScope:
modelscope.cn/models/MiniMax…
API: platform.minimax.io
M Plan: minimax.io/m-planMiniMax
MCode for desktop: agent.minimax.io/download
claude wrote machine-checked proofs that 239 fuzz entry points across 123 open source libraries can't crash on inputs up to 64MB, no brute forcing needed. proofs and the skill to run it on your own repo are public, i'm pointing it at my deps
New on our Frontier Red Team blog: using Claude to formally verify software at large scale.
Formal verification can prove that code is free of certain bugs, but historically it has only been feasible for small codebases (or required a lot of expert time).
With AI models, however, this is beginning to change. In collaboration with @theoremlabs, we used Claude to prove security properties in over one hundred real-world open-source libraries over the span of a few weeks.
Read our research in full here: anthropic.com/research/forma…
nvidia built a thing that watches every gpu, cable and job in a datacenter and points at the one part slowing the whole training run, so an event storm becomes one incident. the hot gpu sitting there degrading quietly while everything else looks fine is the whole problem
We’re launching NVIDIA Minerva to bring end-to-end observability to AI factories at scale.
The NVIDIA Minerva AIOps platform connects infrastructure and workload telemetry across GPUs, networking, storage, and schedulers, using AI to surface root causes and reduce alert noise across factories with hundreds of thousands of GPUs.
Read the news from #NVIDIAGTC Berlin: nvda.ws/4yD3tbb
isaac lab 3.0 is out: nvidia's free sim for teaching robots to move, and you now pick a different physics engine for each task instead of one. congrats to whoever shipped it
We’re excited to announce the official release of NVIDIA Isaac Lab 3.0. 🎉
Isaac Lab 3.0 brings a new modular and lightweight architecture along with major new capabilities across the robot learning workflow: multi-physics and multi-rendering support, training at scale, and streamlined deployment on real hardware.
With Newton integration, the new OVRTX renderer, RL post-training, and LEAPP deployment, there’s a lot to explore.
Here’s a look at what’s new 🧵
(1/7)
a "skill" is just a folder of instructions that teaches your agent one job, and anthropic handed the whole standard to a neutral foundation instead of keeping it. now it carries over between chatgpt and cursor. congrats team, right way to give something away
We're donating Agent Skills to the Agentic AI Foundation (@AgenticAIFdn), the same foundation we donated MCP to.
Since December 2025, 70+ agent platforms have integrated the standard, including ChatGPT, Cursor, GitHub Copilot, and VS Code.
Here's what that means for you:
openai shipped an api for the small calls an app makes all day: is this spam, which team, how bad. no essay first, ~10x faster, up to 400 per request. the boring piece i'll actually use
hassan shipped hallmark v2: you drop it into claude code / codex / opencode and your agent stops making the same purple gradient page as everyone else. 48 themes, and it opens the page in a real browser and calls out the slop itself. installing tonight
modal open-sourced the training server itself. your existing script runs unchanged, and you pay for gpu hours instead of per token. an api losing to a github link, love it
Tinker changed how people do post-training. We're big fans, and so are many of the teams we work with. But as those teams scale up, they keep raising the same two problems: per-token pricing gets expensive quickly, and you’re limited to the models and features the provider decides to support.
We think a good API deserves more than one implementation.
Today we're releasing Spindle, an open-source, Tinker-compatible training server that runs in your Modal workspace. Your existing Tinker scripts run unchanged, and because you own the server, you pay for GPU time instead of tokens.