In agent driven startups, you can scale with vendors like Replit. I patched up 40+ apps < 2 hours! There is a page to view statuses of all of the vibe coded apps & prototypes. Where you can kickoff security patches and fixes and then deploy! https://t.co/7aFZeF5Bhw

@jurlycat shows Phone Harness, an open-source tool that lets Claude Code and Codex control a real phone by reading the screen, tapping, swiping and typing. Per the post it works on iPhone through iPhone Mirroring and on Android through ADB, with nothing installed on the phone and a Hermes plugin too. Existing mobile apps becoming a surface for agents is a big unlock. https://t.co/KnjJu4Gbfz

@ArgusPlatformAI On the MATH (eval) dataset the tiny 0.6B student got 44% accuracy after distillation (up from 15% for the base model). The 0.6B reasoning reference model got around 48%. The 671B teacher model got 91.2%. But yeah, it's also 1000x larger.

@NicolasZucchet @heyyalexwang Transformers work completely different from NN. Memorisation is also a very unfortunate term. Conditioned token sampling that matches a sampled sequence up to a certain threshold seems a better phrasing. Can’t really make any sense of your terminology. There’s no memory anywhere.

Spent a bunch of tokens over the last few weeks building a fully procedural New York simulation. https://t.co/YSrGNoIYsK It's only 784KB gzipped and the code is on github, so you can do whatever you want with it ✌️ https://t.co/tj5wXVxhDR

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: https://t.co/zUyx8CjBEk

Introducing Voxtral 3: state-of-the-art Transcribe and Speech models powering Mistral’s new voice agents API. 🧵 https://t.co/YA9HNVfq0u

New paper: AI evaluation is a burgeoning field. How can researchers entering the field understand best practices and open problems? In our paper, led by @flxcn, we develop a concrete list of open questions and best practices for agent evaluation. We also share recommendations for evaluators, AI companies, policymakers, and more.

This user gave Perplexity Computer, SSH access to a Raspberry Pi 5 over Tailscale. Overnight, it autonomously turned it into a game console with an arcade UI, toolchains and MCP integration. The console generates GBA, C, Rust and terminal games with automated compilation, playtesting, and error fixing. Computer also built a custom USB controller service with remapping, turbo, rumble and per-game profiles, plus a Minecraft-inspired multiplayer voxel game in C (141 KB binary) where the user can play alongside Computer as an AI-controlled robot, directed through in-game chat via MCP. https://t.co/hHzw2mO1Nu

"Create a simple 3D animation from this video." An LLM agent builds a scene and uses our FlowHMR for mocap. The agent then handles retargeting, motion fixes, camera setup and rendering. Finally, MiniMax-H3 turns the 3D render into photorealistic video. https://t.co/KMHZNWTSHW https://t.co/plKShpGsLQ

GPT 6.1 Terra is a strong frontier model baseline for document parsing. ✅ It spikes at table parsing, coming in 5th on our overall leaderboard, ahead of 6 Sol and behind Opus 5.5 / Astra ✅ It is decent at content faithfulness and reading order ✅ It struggles on layout, charts, and formatting Coming in at ~1c per page, it's not a bad option for table-heavy pages! You can use these alongside specialized OCR solutions (like LlamaParse) that will give more consistent performance across a variety of dimensions. Check it out on ParseBench: https://t.co/PWczfhp0OX

In collaboration with @HuggingFace, we’ve released the largest open robot dataset, alongside their first open System 1-System 2 model that can reason over step-by-step actions. FineART is built with 40K+ demonstrations across 151 tasks and 500K subtask labels. Together, the dataset and model were designed for each other, and both are available on @LeRobotHF where robotics developers work today.

An update on our evaluations of AI agents conducting open-ended AI research. Most evaluations of agents conducting AI research focus on narrow, verifiable tasks. But AI research is often open ended. In July we tested the ability of frontier AI agents to conduct open-ended AI research, using Claude Opus 4.8 and GPT-5.6 Sol. We gave agents research questions from two unpublished papers and thousands of dollars of API credits and compute. The authors of the original papers reviewed the AI-generated papers, following a NeurIPS-style template, and rejected both papers. Since then, OpenAI has released GPT-6 Astra and Anthropic has released Fable 5.1. Can these new models conduct open-ended research with a $10,000 budget? In follow-up experiments on two papers (one with Fable 5.1 and another with GPT-6 Astra), we observed meaningful improvements to the agents' capabilities at experimental design and resource management. But we also saw continued limitations in creative problem solving, judgment, calibration, and communication. As a result, the expert reviewers still unambiguously rejected the agents' papers. We plan to continue running these evaluations with better models, better scaffolds, and higher budgets. If you are interested in long-running agent evaluations, reach out!

🎥 A long video. One coherent 3D scene. Meet HiP3R: piecing together local 3D reconstructions to keep the scene consistent across thousands of frames. Watch the demo and explore the details 👇 Project: https://t.co/BzGoFauLfq Paper: https://t.co/ttWFBtkbCc Code: https://t.co/33tRzAGuTp

We’re open-sourcing Drex-Max v1 The most capable Multimodal Decision Model ! And yes: it runs 100% locally on Apple Silicon (M4 & M5) Enjoy Building with Nace! Cloud Inference: https://t.co/fm3GXfjaqx Local Inference: https://t.co/9nwcBn7c2D https://t.co/Wgu8P4zByR

+1 more

This is really cool work on our synthetic hospital benchmark. For patient diagnosis, this shows that small models + SFT + the right tools > GPT-6 Luna. Interesting is that the tools help small models, but not big models. So gaps in capability can be replaced by tools

MemT-1.0: Memory-native foundation models MemTensor introduces a new architecture where persistent memory is part of model computation, enabling 1.86x inference speedup and 66x storage reduction at 128K history. https://t.co/kU21sHGSPE

S4VY is an incredible advancement in the field of real-time 4D joint segmentation and depth prediction! We’ll be releasing the code and checkpoints soon, so stay tuned for more updates.

Open weights models fall further behind in v5. The leader, Kimi K3.1 Pro (weights not yet released), scores 46, 12 points behind the leader, compared with a 3-point gap in v4.3. Of models with released weights, DeepSeek V4.1 Flash 1017 (max) and GLM-5.3 (max) score highest at 32, and Claude Haiku 5.5 (max) outscores all of them. The gap is widest on HLE-Diamond and Terminal-Bench-Science, where Kimi K3.1 Pro scores 37% and 34%, compared with 68% for Claude Opus 5.5 (max) and 63% for GPT-6 Astra (max).

Drop 1/15: Introducing Joy, a dual-channel conversational dataset. Real, unscripted conversations between two people, each on their own track. Built to teach voice models how people actually talk: when to listen, when to jump in, when to say “hmm”. 100,000 hours. 50,000 speakers. 10 languages.

wtf is a "decision model"? Like architectural details, not hand-wavy explanations. Come learn from @barrowjoseph with me. https://t.co/JbixiT9oyN

One broadcast clip → solved the camera → reconstructed the trajectory → physics for every thruster pulse → a WebXR replay you can step into. I recreated the Roadster ramp stunt with IWSDK. Try it: https://t.co/dfXytTHyA7 Code: https://t.co/2X4dcnOxWF https://t.co/MgTqdE6Joa

Silt is a design canvas wired into your real codebase. Each project is a repo, so your design system, tokens and code conventions sit in the agent's context while you design. That one idea changes everything the agent can do for you. https://t.co/93xQckyfSW

Wow, diffusion models (used in AI image generation) are also game engines - a type of world simulation. By predicting the next frame of the classic shooter DOOM, you get a playable game at 20 fps without any underlying real game engine. This video is from the diffusion model. https://t.co/p13fSVvl7K

Introducing Gemini 4 Carbon, it's a big jump over Argon which we released just 2 weeks ago. Gemini 4 Carbon is SOTA on cyber, long horizon tasks, ranks #1 on Artificial Analysis V5, significantly improves on usability in domains like coding, and is more aligned. https://t.co/Ikk9YedjCQ

Purdue University's IDEAS Lab built an autonomous drone that tracks runners through dense forests completely without GPS or cloud networks. By running 3D LiDAR mapping and obstacle-avoidance AI entirely on-chip, it can weave under tree canopies and predict target movement through blind spots in real time. Massive implications for next-gen search & rescue and edge AI. https://t.co/HFsnKweHLY

We placed the fruit fly brain inside Odyssey-3, a frontier world model. The lil’ guy is now flying around totally imagined, simulated worlds all by itself 🫶 https://t.co/lvKrGoTtKr

Finally summarized what I've been sending to friends in DMs https://t.co/rnf3M16NzI

Meet NVIDIA Nemotron 3.5 Transcribe 🎙️ An open-weight, promptable 30B MoE model with ~4B active parameters that brings transcription in 40 languages, speaker attribution, translation, contextual prompting and audio understanding into one checkpoint. Watch it keep track of who said what when everyone’s talking at once. 1/ 🧵

AI may well be able to write massive software projects “from scratch”, but that almost certainly isn’t the fastest / highest performance / highest security / most token efficient way to do things. Good building blocks matter. In addition to RL’ing the agents on final task loss, we should try to get them to regularize the building blocks as well. What would the standard library look like that makes the agents most effective? It is likely at the limit that it turns out to be some alien organization of code, but given the existing human-bias in training, I bet that in the near term it would be lovely code that old programmers would appreciate. I suspect the same could be said for mathematics, but that is out of my lane.

GPT-6.1 Terra continues to advance the cost-intelligence frontier across our model families. It’s a significant step up over GPT‑5.6 Terra, with 50% lower standard API prices. In ChatGPT Work and Codex, your usage allowance goes further with GPT‑6.1 Terra than with GPT‑5.6 Terra—and far more than with GPT‑6.1 Sol or GPT‑6 Astra. For in-depth research or complex coding, choose GPT-6.1 Sol or GPT-6 Astra.