GPT-6.1 Terra: delivering fast and capable computer use at a lower price. GPT-6.1 Terra understands what’s on screen, takes action across apps, writes naturally, and is a more capable model for your whole workday with the efficiency to take on more of it. https://t.co/8JTN1yonmV
#Product updates#Agents#OpenAI
Meet the world-class research team behind SKY-001, the first Engineering World Model. Solving complex design and engineering problems in the physical world is orders of magnitudes more difficult than LLM agents using next token prediction. Our team at @skyfallai decided to take a different path than everyone else out there in order to achieve the research breakthrough. Last week, we shared how SKY-001 stacks up against frontier LLMs, showing the clear advantage World Models have on complex real-world tasks like engineering design. Here is a sneak peek video where you can learn directly from the researchers about what makes our approach unique, what training and data collection paradigms were used, and why we believe World Models are the next frontier in AI. I’m incredibly proud to work alongside this team.
#English text is being prepared.#Multimodal#Reasoning
#Product updates#English text is being prepared.#Agents
English text is being prepared.
We’re excited to open-source Hermit! 🐚 A harness evolution framework for continual learning and RSI: train reusable adapters without changing your model or harness. Multiple methods, one framework. GitHub: https://t.co/ateUcBiB8Y #HarnessEvolution #ContinualLearning #RSI #Agent
#Papers/Research#Product updates#English text is being prepared.
#Safety / alignment#Agents#Open-source ecosystem
#Product updates#Agents
Nace AI has released Drex for Search, applying its Drex decision model to document retrieval. Coding agents such as Claude Code, Codex, Cursor, and OpenCode can invoke it via CLI, SDK, MCP, or REST. Paired with GPT-5 on BrowseComp-Plus, it achieves 92.05% accuracy with 12.61 search calls per question, 16.6 times fewer than the top-ranked entry. On VisDocAgentBench, using a Qwen 27B planner, it achieves 77.5% R@1, 10 percentage points higher than Claude Opus 5.
#Product updates#RAG#Agents
The open-source project nxnative can convert users' own Switch games into native apps for iPhone, iPad, and Mac, with the games' original code running directly on Apple chips without going through an emulator. Its GitHub page already lists more than 60 tested games, including Mario Kart 8 Deluxe, Metroid Dread, and Super Mario Odyssey; some games, with specific updates, can go from 30 fps to 60 fps. Using it requires providing your own game files and decryption keys extracted from your own console, and running build scripts on a Mac with Xcode installed; the finished product must be sideloaded via tools such as AltStore or Sideloadly, and it only supports iPhone 15 Pro and newer models or iPads with M1 and newer chips, with the system requiring iOS 26 or iPadOS 26, and it does not support amiibo or online multiplayer.
#English text is being prepared.#English text is being prepared.#English text is being prepared.
The team of Florian Euchner at the Institute of Communications and Navigation of the University of Stuttgart has released the open-source firmware ESP-SDR, which uses an undocumented path in Espressif's test library librftest to let ordinary ESP32 boards receive raw 2.4 GHz RF signals without additional hardware. It supports the classic ESP32, S3, C6 and the C5, which can cover 5 GHz, but due to USB bandwidth limits most models can only take short snapshots, making them better suited as spectrum analyzers and unable to replace the roughly 30-euro RTL-SDR receiver dongle. The firmware is released on GitHub under the GPLv3 or later license and offers a browser installer and the ESP-WebSDR web interface; the team deliberately does not provide transmit functionality.
#English text is being prepared.#English text is being prepared.#Open-source ecosystem
Open-source security company ProjectDiscovery has added Neo Triage to its Neo security testing platform. It can reproduce each report against the actual asset where the vulnerability resides in an isolated sandbox, and issue conclusions such as valid, duplicate, or junk, while attaching test traffic as evidence. The feature targets the vulnerability backlog caused by AI scanning and code review. The company says manual reproduction of a single item takes 15 minutes to 4 hours, and manually processing a backlog of 100,000 items would take about 50,000 hours, while reviewing Neo verdicts takes about 8,300 hours. Neo Triage is now available to all Neo plans, including the free version.
#Product updates#Safety / alignment#Agents
Microsoft will add AI security features to Teams that analyze conversation content to identify phishing, scams, and identity theft, with a focus on new contacts. The feature no longer only checks malicious links and files, but instead identifies suspicious behavior in messages; Microsoft says messages will not be stored or used to train models, nor used for monitoring or advertising. The feature will begin rolling out in early November 2026 and is expected to be fully available in the first half of December; IT teams will not need to configure anything, as it will be enabled automatically.
#Product updates#Safety / alignment#Microsoft
Canva has released its first self-developed AI image model, Canva Craft, now available globally to all Canva users and focused on designer scenarios. The model uses a two-stage architecture and, combined with Canva's prompting tools, can plan the layout before generation; its training data comes from Canva's asset library, licensed and public databases, and user data from those who opted in to training. On the DesignArena image model leaderboard, Canva Craft ranks fourth, behind only OpenAI's ChatGPT Images and ahead of Google's Nano Banana and Microsoft's MAI-Image.
#Model releases#Image generation#Google
At PyTorch Conference 2026, AWS and Google Cloud successively open-sourced their respective PyTorch backends for their AI chips: TorchNeuron for AWS Trainium and TorchTPU for Google Cloud TPU. Both were previously available only to select users in private testing. AWS says TorchNeuron requires only switching to .to('neuron') to run, while maintaining compatibility with eager execution, torch.compile, and standard PyTorch APIs; Google Cloud says TorchTPU supports out-of-the-box model execution and TPU migration. Both aim to lower the porting barrier for running PyTorch outside of NVIDIA CUDA.
#English text is being prepared.#Open-source ecosystem#English text is being prepared.
Replit has released Replit for Business, offering enterprises a foundation that lets agents use company knowledge and systems within controlled boundaries, with the goal of helping enterprises transform into self-driving companies. Replit says that in the first half of 2026 its engineers' code output nearly tripled, while review time, rollbacks, and product incidents remained flat. Customer SMFL says a development process that originally required 6 people and 6 months was completed by 1 person in 1 week, and full lifecycle costs dropped from about 3 billion yen to under 1 billion yen.
#Product updates#Agents#English text is being prepared.
Cognition announced an upgrade to Devin Security, adding deep scanning driven by frontier models such as Claude Mythos to discover complex chained vulnerabilities in large codebases. The product will build a continuously updated threat model together with teams, reorder vulnerabilities by exploitability and potential damage, and can reproduce, remediate, and test in parallel in isolated virtual machines, submitting PRs with evidence. Cognition also committed hundreds of millions of dollars in Devin credits to provide free scanning and remediation quotas for existing and future enterprise customers.
#Product updates#Safety / alignment#Agents
Introducing Drex for Search 🔍 Supercharge your agentic search with Decision-powered retrieval Standard file search holds AI agents back. Drex changes that: • 62% fewer steps to find answers • 56% tokens saved • +4% performance boost across Claude Code, Codex, OpenCode, Hermes, and PI Validated across 6 professional work benchmarks like OfficeQA (529 real-world queries) Upgrade your agentic workflows with the File Search API Try for free: https://t.co/vdEavE1dsd Faster agents. Cheaper runs. Sharper accuracy.
#Product updates#RAG#Agents
#Product updates#Safety / alignment#Agents
#Product updates#Agents#Open-source ecosystem
MiniMax M3.1 Flash: Space Performance + Bunny Speed 🚀🐰 Yes, Space Bunny - which has topped global leaderboards with 200T+ tokens - was an earlier version of M3.1 Flash. 💰Redefining the performance economics of agentic AI. M3.1 Flash achieves frontier-level intelligence across key tasks such as Cowork (Automation-Bench: 62%; ALE: 30%), Coding (DeepSWE: 74%) and Visual Reasoning (BabyVision: 91%), at 1/3 the cost of M3 and 3x the speed. API pricing (flat rate): $0.15 / $0.60 per 1M input/output tokens, with cache reads at $0.015 / 1M tokens. Model weights will arrive on Oct 23. API: https://t.co/fHRdSV73Hr M Plan: https://t.co/XfJWSyauda MiniMax Code for desktop: https://t.co/SGmLTI2HQm
#English text is being prepared.#English text is being prepared.#English text is being prepared.
#English text is being prepared.#Agents#English text is being prepared.