Mistral Studio just rolled out a major upgrade, transforming the platform into an end-to-end hub for building and hosting AI applications. Developers can now wire up reusable knowledge connectors, deploy reasoning agents and deterministic workflows with a single command, and turn production run traces into evaluations to continuously refine their apps.
SpaceXSI is rolling out Slack integration for Grok Bot, letting users connect bots individually to workspaces. Once connected, you can DM or @ mention bots directly in Slack while keeping all plugins, context, and memories intact. The feature is slated for the macOS 1.1.0 build.
Anthropic is working on a collaborative mode for Claude Code Projects. The platform's Projects page now summarizes recent thread activity, while Anthropic's Lydia Hallie teased yesterday that multiplayer functionality is in the works. Why it matters: Adding real-time team collaboration could make shared AI coding workflows far smoother.
Lattice just launched a local LaTeX workspace built for researchers, packing writing, PDF previews, and citations into one desktop hub. It features two-way sync with Overleaf and connects directly to coding agents like Claude Code to fix build errors and refine papers, cutting out constant app switching.
Anthropic just upgraded Claude Code cloud sessions for Pro, Max, and Ultra subscribers. The AI can now automatically figure out project dependencies, network access, and required API keys to spin up and run code in the cloud with minimal manual setup.
Claude for iOS is testing another interface redesign that puts Chats, Projects, and Code into separate tabs along a side navigation bar. Other options like Dispatch and Personal Finance have been tucked away into the profile menu, shifting away from the previous bottom bar layout.
OpenAI just teamed up with the Frontier Model Forum and member firms on a new framework to identify when bypassing AI safeguards triggers serious cyber harm. The goal is a shared standard to measure unlocked capabilities and ease of misuse, helping security teams prioritize where to build stronger defenses.
Anthropic just shared a look into how AI is transforming scientific research. Instead of researchers manually writing and weighing competing theories, AI can now write, run, fit, and tweak models as code to rapidly uncover the explanations backed by actual data.
Maket just rolled out Text to Furniture, a new tool that turns text prompts into 3D furniture models. You can describe specific pieces like an oval burl wood dining table, and it creates a scaled 3D model that drops right into your floor plan.
OpenAI teamed up with biotech firm Ligo to launch BioBasicsBench, evaluating how AI agents interpret raw data across 100 routine wet-lab tasks. Running at maximum reasoning effort, GPT-6 Astra scored 66.7%, comfortably beating GPT-5.6 Sol's 43.8%. Still, the models struggle with control checks and colony counting, showing AI lab assistants aren't ready to fly solo just yet.
Meta just rolled out new specialized models trained on Muse Wonder to its Muse and Meta AI apps. The update brings Active Memory and Active Focus to all users, cutting critical info slip-ups by 2.4x and boosting everyday task completion and self-correction by 67%. Basically, you'll repeat yourself less and hit fewer dead ends.
Nace AI just released a 51-page technical report on Drex, its open-weight model family designed to pick the best answer from fixed options. Trained on 1.15 million verified decisions, it delivers calibrated probabilities in just 183ms. Open-weight versions are available on Hugging Face alongside a web test console.
Meta just rolled out Spaces inside its Muse app, letting users spin up personalized interactive trackers and boards simply by describing what they need in chat. The spaces pull data from connected apps to update in the background and can be pinned or scheduled across your tabs. Collaborative sharing is coming next, backed by specialized models trained on Muse Wonder.
Meta, Google, and TikTok just filed federal lawsuits to kill California's AB 1709 before it kicks in on Jan. 1, 2027. The law bans autoplay and personalized algorithmic feeds for users under 16, threatening fines up to $50K per affected child. The platforms argue the rule violates the First Amendment, setting up a major legal showdown over how far states can go in regulating algorithm-driven content.
Amazon just added a text messaging feature to its Alexa for Shopping assistant, allowing US shoppers to order groceries and household items simply by texting a contact named “Amazon Alexa.” Users can describe items or send photos, and the AI agent searches Amazon listings, recommends matches, and completes purchases on their behalf. It is currently rolling out to an invite-only group before expanding to millions more.
Cactus just dropped Mirage, a tiny 51.3M-parameter vision model compressed to roughly 22MB via 2-bit quantization. It handles scene description, object detection, OCR, segmentation masks, and spatial reasoning in a single pass—clocking 74ms on a Galaxy S26 NPU. It proves full-featured visual perception can comfortably live on device edge chips without touching the cloud.
Video Rebirth just released HyperFlow H3 Flash, an open ~14B distillation of MiniMax-H3 that generates synchronized 768p video and audio in 4 to 8 steps. The Q4 GGUF weight drops down to 9.1 GB to run on 24GB RTX GPUs, while 4x Blackwell B200s achieve playback-speed real-time generation. Why it matters: Cutting-edge, synchronized audiovisual models are finally breaking out of hyperscaler clusters and onto local creator hardware.
Physical Superintelligence (PSI) just settled the 18-year-old Aaronson-Ambainis conjecture using a research system powered by an unreleased Anthropic model. The entire proof was machine-verified by the Lean theorem prover in just three days. Why it matters: The newly minted theorem mathematically proves quantum computers cannot achieve exponential speedups on unstructured random data, setting a hard theoretical ceiling on quantum hype.
Anthropic just tested its Claude models inside the brutal economies of EVE Online and EVE Frontier. Models like Fable 5 tripled earlier returns and grew initial capital by up to 600% through disciplined spread trading, but in live four-agent co-op missions, the team completely failed to catch a covert saboteur. Why it matters: Frontier models are getting scary good at mechanical market making, yet multi-agent oversight and trust remain wide open vulnerabilities.
Anthropic just unveiled a new agent memory framework for Claude inspired by human sleep, dubbed 'dreaming.' During idle time between sessions, agents review raw logs to correct faulty assumptions and distill broad strategies rather than hoarding every trivial detail. Why it matters: smarter agents don't just remember more—they know what to prune.
Anthropic AI for Science grantees just outlined a roadmap for how AI can automate scientific discovery instead of merely curve-fitting. Drawing on the classic Hodgkin-Huxley neuron model, the researchers argue AI agents must generate distributions of plausible models via simulation-based inference. It marks a push to turn AI from a black-box calculator into a tool for genuine scientific understanding.
OpenAI just announced it will stop using non-disclosure agreements with public officials for future data center projects, following voter pushback over rapid tech expansion. The company also pledged to cover the full cost of its electricity and related infrastructure, while publishing local job creation data. Amazon made a similar NDA pledge in October.
Google just open-sourced Agile Deliberation, an AI framework that flips traditional prompt writing. Instead of forcing you to define subjective concepts upfront, the model actively hunts down tricky edge cases to help you pinpoint your own criteria—making content moderation and brand style labeling much cleaner.
A new report from safety nonprofit METR estimates Anthropic’s model R&D is moving nearly 2x faster, with AI driving about 80% of the jump. The evaluation also teased unreleased internal models like Model 4, suggesting the recursive AI self-improvement loop is already quietly running inside the lab.
Serval CTO Alex just announced five new roles across San Francisco, New York, London, and Austin, offering new hires a paid luxury vacation worth up to $10,000. The lavish recruiting move follows a $150 million funding round led by Sequoia in late October that bumped the company's valuation to $3 billion.
Serval is recruiting engineers to build Catalyst, an AI agent that automates IT technician workflows from natural language prompts. Teams can test and approve the agent-generated code before scheduled execution. To lure talent, the startup is offering every new engineering hire a weeklong, $10,000 paid vacation before their first day.
Meta just rolled out realtime voice and expressive avatars to its Muse app across mobile, web, and Mac. The avatars react with subtle expressions matching your conversational tone, support natural mid-sentence interruptions, and feature hundreds of voice personalities. Currently live for US users, with support for AI Glasses coming next.
Yukon proved capable of keeping table latency under two minutes during a one-hour, 1M events-per-second benchmark run. It saves tables in open formats directly to your cloud storage and logs ongoing changes, allowing AI agents to inspect what updated without scanning whole tables.
Eon just introduced Eon Yukon, a serverless data engine designed to turn streaming events into clean Iceberg tables for AI agents within seconds. At 1M events per second, it uses 4x less compute than Spark and 7x less than Flink while exposing live stats through MCP. It basically cuts out bulky data pipelines so agents can finally act on fresh context without massive infrastructure overhead.
Perplexity just started rolling out Teammates, its new proactive agents, to Perplexity Max subscribers on the web. They operate in an isolated environment with memory, tool calling, and the ability to execute computer tasks. Mobile app support is coming soon.
OpenAI just expanded access to its Daybreak security platform for organizations of all sizes, independent researchers, and open-source maintainers. The program includes Daybreak Blue for defensive validation with fewer refusals and Daybreak Red for authorized testing, alongside a broadened GPT-6 Astra alpha rollout.
Agility Robotics just listed on Nasdaq via a SPAC merger at a $2.5B valuation, making it the first U.S. humanoid pure play. But with just $1.8M in 2025 revenue and over half coming from a single investor, Digit must now survive the scrutiny of quarterly earnings to prove warehouse humanoids can turn a profit.
Ant Group just unveiled Ling-3.2-flash, a 560B-parameter native multimodal MoE model that only activates 25B parameters per token. It natively processes text, image, and video inputs across a 1M context window, handling tasks from multi-agent workflow coordination to 3D spatial reconstruction.
Luma AI just revealed its new Ray 2 video model, extending generation limits from 5 seconds to up to a full minute, with outputs ready in as little as 10 seconds. The startup is also bringing its visual models to Amazon Bedrock and tapping AWS custom silicon to slash compute costs.
Anthropic just rolled out Personal Finance in Claude, launching in beta for US paid subscribers on Pro, Max, and Ultra plans. Users can securely link bank and card accounts via Plaid to ask plain-language questions about balances and spending habits. Access is strictly read-only with zero ads, and chats in the finance hub are excluded from model training.
Runway just halved the hardware footprint for its 57B-parameter Characters model from eight H100s to four while sustaining 24fps video. By moving FP8 quantization and RoPE to cache write-time and fusing communication kernels into matrix multiplications, it brought diffusion steps to 147.4ms. That makes serving live, interactive video avatars far more viable at scale.
Google just launched Gemini Omni 1.2 Flash, boosting native video generation to 15 seconds while slashing pricing by 30% to $0.07 per second. The update adds multi-turn conversational editing so you can tweak scenes without regenerating from scratch, alongside tighter reference consistency and realistic lip-sync. Google account users can now generate three 15-second clips every month in Google Vids for free.
Thinking Machines just unveiled Inkling 2 Small, a 594B-parameter MoE model with only 17B active parameters built for autonomous agentic coding and visual reasoning. It can run experiments, fix bugs, and refine interfaces using visual feedback in the loop. Full weights are slated to drop under Apache 2.0 next week, but you can already test it on Tinker Playground.
Open-source project hunch turns Google's EmbeddingGemma 2 into a local, reflex-speed decision engine running directly on commodity CPUs. It acts as a fast student that learns from frontier LLMs and takes over repetitive chores like tool trace verification in just 22ms. Basically, it lets you slash agent inference bills without needing a GPU.
Oumi just launched an agent-building platform powered by FAiE, an AI engineer designed to handle everything from training to verification. Users can simply correct mistakes in chat, prompting FAiE to update datasets, retrain the model, and measure gains in roughly 15 minutes instead of months of engineering.
OpenAI just launched GPT-6 Cyber, a specialized cybersecurity model designed to find, exploit, and patch software vulnerabilities without wandering out of scope. In a single automated run, it validated and patched 948 Linux kernel privilege escalation bugs at a $1.2 million API cost. It is currently available to select Daybreak Red users.
Coding agent platform Conductor just joined Meta to build what its team calls the next interface for developers. Working alongside MSL, they plan to launch something new early next year. Why it matters: Meta appears to be assembling the pieces for its own AI coding super app.
Google's Antigravity just rolled out v2.29.0, adding a voice mode powered by Gemini Live so users can hand off requests to agents by speaking. The update also introduces cloud-run remote agents for Google AI Ultra and Pro subscribers. The rollout is rolling out gradually to all users.
Kimi just released the open weights for K3.1 Pro on Hugging Face under a dedicated license. Billed as its most capable 3T-class model yet, the release packs native multimodality and agentic capabilities directly into the hands of open-source builders.
Bayworks just launched an app that lets dealership technicians dictate their repairs out loud and automatically format reports for automaker warranty claims. The app prompts mechanics if required measurements are missing, creating an audio audit trail for every fix. The YC-backed startup has scaled to 80+ dealerships and 4,000 daily repairs in 110 days.
ChatAndBuild just launched Living Brain, a persistent memory layer that lets ChatGPT, Claude, Cursor, and custom agents read and edit the same context. Users can switch between AI apps without re-explaining background details. It boasts a 464ms median retrieval latency and up to 22.6x fewer tokens per query.
Nace AI just launched a 120% cashback promotion for teams switching to Drex Max, its decision model for AI agents. Developers can migrate by changing three environment variables and earn $1.20 for every dollar spent. Nace claims 1,000 browser tasks on MiniWoB++ cost $2.80 with Drex Max, compared to $31 with Claude Code and $25 with Codex.
ClickUp just started rolling out Brain² browser use to its Brain assistant. For sites without official integrations, the AI can now click around, fill out forms, gather info, and download files straight into ClickUp—with live viewing, step logs, and full session recordings included.
OpenAI just announced it will acquire Pierre Computer Company to upgrade Git workflows and code review inside Codex. Pierre's Trees and Diffs libraries already power file navigation and change reviews in the Codex app, and its open-source tools will remain open. Why it matters: as software teams deploy multiple coding agents in parallel, OpenAI wants to redefine modern version control for the agentic era.
Nous Research just opened a waitlist for its new Hermes Mobile app after initial beta spots filled up almost instantly at NousCon 2026. The team says it is actively working to free up more beta slots while polishing the agent app for a wider public rollout.
Tencent just released Hunyuan Hy4, upgraded with vision and computer-use capabilities to run complex workflows across apps like Blender and CAD. It tackles knowledge work like financial modeling and 3D scientific reconstruction, with open weights coming in days.
A two-person lab called Linum just open-sourced Pyramid-JiT, a 2B-parameter text-to-image model trained entirely from scratch for only $21,000. It ditches the standard VAE to generate images directly in pixel space, delivering solid visuals on a shoestring budget. While it struggles with text and complex anatomy, the Apache 2.0 release proves training diffusion models doesn't require millions.
Pangram just launched its flagship Image 1.0 and Image Flash 1.0 detection models, built to spot synthetic images and videos from frontier generators. Image 1.0 outputs heatmaps that highlight exact manipulated spots, while Flash 1.0 runs frame-by-frame video analysis with 98.7% accuracy on the DF26 benchmark. Both models are available via API and a Chrome extension for real-time feed checking.
Ricursive Intelligence just released Ripple-65B Preview, a dense open-weight model with a 1M context window trained for $32M. Skipping MoE complexity, it scored 80.52 on TerminalBench 2.1, beating massive frontier open-weight models over ten times its size in agentic coding.
OpenAI just launched Images 2.6 across ChatGPT, Codex, and its API, boosting generation speeds by 50% and adding native support for animated GIFs. Developers get two API variants—including the six-second Flare model—plus SVG exports and fine-grained scene complexity controls. It is live now for all subscription tiers.
Anthropic's biology lab just mapped the first 3D structure of a vault particle from prokaryotic cells, using Claude at every step of the research. Claude helped search protein databases, design DNA constructs, and process cryo-EM data—shrinking a weeks-long modeling process down to 34 hours. Why it matters: Vaults have baffled biologists since 1986, and this shows AI can meaningfully speed up real-world wet-lab discovery.
Uber just revealed on its Q3 earnings call that out of roughly 320 million trips and orders during Halloween week, only four rides were booked via AI agents, with Uber Eats logging just 500. CEO Dara Khosrowshahi noted app booking takes just 50 seconds anyway, delivering a reality check to the agentic web hype.
Tencent just introduced Hy4, its new flagship model rivaling top open models on benchmarks with built-in vision and computer use. The model is already accessible on Tencent Cloud and OpenRouter, with open weights dropping soon alongside ecosystem partnerships with vLLM and SGLang.
Anthropic just open-sourced FlashProve, a provable inference system using cryptographic zero-knowledge proofs to verify an AI response actually came from the claimed model. Verifiers do not need weights or GPUs, checking proofs on a standard CPU in seconds, though generating proofs still takes 72 to 94 times more GPU compute.
Grok Bot just rolled out a new shopping experience, letting users hunt for the best deals and track ongoing orders directly in the chat. It is also adding passkey support for purchase approvals soon, basically turning the assistant into an end-to-end shopping agent.
Envato just launched MonoDesk, a workspace tailored for creative freelancers and boutique studios. Its built-in AI assistant spots out-of-scope client requests and drafts price quotes automatically, while turning accepted proposals into trackable project roadmaps. It also links with Claude and ChatGPT to log hours, and is completely free during public beta.
OpenAI just revealed it's working with Chip Ganassi Racing engineers to turn 3D scans into editable CAD models and manufacturing drawings. The company is also collaborating with Mazda to explore new vehicle designs through simulation, relying on partner engineers to review outputs and sharpen the models.
Apple just introduced always-on "Audio Intelligence" features in the watchOS 27.3 developer beta, bringing conversational AI directly to your wrist. A double-press of the Digital Crown pulls up a 15-second text rewind of what was just said, while Siri Recap automatically jots down key points and titles. Apple claims raw audio is processed strictly on dedicated hardware and never saved as a file.
Netflix just launched an experimental 'AI Discover' tab on its TV app, serving up themed picks based on your mood—like 'Something Light' or 'Quick Watches.' Powered by a new intelligence layer, it remembers titles you reject and tunes suggestions in real time. It marks a shift away from endless grid scrolling, rolling out first to English-language TV users in the U.S.
Selendis just launched Archon, a video foundation model that automatically turns raw footage into finished edits in minutes. Built with musician Steve Aoki, it learns a creator's taste from past videos and reference clips to get sharper with every project. Early access is currently open with three months free.
MIT researchers found that OpenAI’s GPT-6 Astra can inspect and directly rewrite neural network weights without any gradient descent or retraining. In one test, Astra modified just 0.016% of a small Qwen model’s weights to eliminate repetitive loops, bumping its math score from 7.9% to 48.1%. If frontier models can reliably program weight space, AI self-improvement might arrive much faster than expected.
Researchers just open-sourced NitroFusion 2, slashing image diffusion down from 8 steps to just 1 or 2 steps. By introducing adversarial distillation directly into the multimodal text stream, it preserves fine details and legible typography even in a single pass. The weights are available on Hugging Face and run locally via ComfyUI.
Researchers from Tsinghua and UC Berkeley just open-sourced uni-robot-agent, a unified harness that lets open-weight models control robots in simulation and the real world. Running locally on Qwen3.8-Flash-Next, it scored 31.39 on RoboDojo, edging past GPT-6 Astra's baseline at roughly 1/20th the cost.
Pruna AI just released P-H3-Image, an open-weights image model built by reworking the MiniMax-H3 video model. By distilling components and slashing diffusion steps from 50 to 8, it runs 5x faster and shrinks down to 20 GB while rendering up to 16 megapixels. ComfyUI already supports it natively.
Sandflow AI just open-sourced sand3d, a Python library built on Blender that lets LLMs construct parametric 3D objects and robotic joints in readable code. The entire API fits in a 3,600-token prompt—using up to 6x fewer tokens than raw Blender script—and launches alongside a massive 100,000-object open dataset.
OpenAI just revealed how its frontier models are tackling physical engineering, with two staffers using GPT-6 Astra to design a hardware prototype in under a week. Internal hardware token usage has jumped 23-fold since February, and partners like Mazda and Chip Ganassi Racing are already testing CAD conversion and aerodynamics simulations. It's not fully autonomous yet, though—prototypes still required human review to fix manufacturing tolerances.
Google just launched Guided Learning globally in Google Classroom and search, while overhauling Gemini and search controls for K-12 students. The education push also includes expanded AI training for teachers and new Google.org funding for classroom tools.
NVIDIA just open-sourced Auto Ontology, an Apache 2.0 framework that builds a semantic reasoning layer over enterprise databases. It translates ambiguous questions into validated, deterministic SQL queries with verifiable data provenance, and plugs directly into agents via MCP.
OpenAI just announced an upcoming internal security sprint alongside Cathedral, putting its own models to work to probe and harden company systems. It is also quadrupling bounties for critical bug findings and pledging $1 billion in subsidized access for frontline defenders over six months. Basically, OpenAI wants to use frontier AI to lock its own doors before attackers find an opening.
NVIDIA just introduced Cosmos Framework and Cosmos NIM microservices to build and deploy physical AI world models for robotics, AVs, and surgical simulation. It features new Cosmos 3 Nano checkpoints capable of generating synchronized multi-camera video and LiDAR data. With speculative decoding, the NIM containers deliver 2–3x faster inference on H200 GPUs.
42AI just exited stealth with a $62.5M seed round to build a transformer-based "behavioral world model" that predicts human choices under market volatility. By turning support interactions and trade histories into psychological trait vectors, the model estimates decision probabilities for 10 financial brokerages. Why it matters: If brokerages can anticipate client behavior before orders are placed, the line between risk management and customer exploitation quickly blurs.
Elon Musk just confirmed that SpaceX slashed Grok Bot usage limits because running the agentic AI has become unsustainably expensive relative to revenue. The sudden squeeze maxed out weekly caps for users over the weekend, proving that even a profitable AI division can get crushed by runaway compute bills. Future subscription limits remain unannounced.
Hume AI just launched its Audio Decision API, an audio-native model that answers custom questions about speech and sound directly from raw audio—no transcripts needed. You can check if a caller sounds frustrated, identify AI voices, or search 40,000 recordings in minutes with simple JSON outputs. Why it matters: instead of relying on words alone, voice agents and call centers can now catch sighs, tone shifts, and unspoken context in real time.
Meta teamed up with Chinese robotics firm Fourier to build Max, a 5-foot, 106-pound humanoid robot designed to handle household chores. Built in just three months and powered by Nvidia’s Jetson Thor chip, the prototype gives Meta a physical testbed to train real-world AI and take on Tesla’s Optimus.
OpenAI just launched its Grid Protectors Program, agreeing to dial back power use at future AI compute campuses during peak grid strain. Its first pilot in Georgia can free up to 1,000 MW of demand response—enough to power 700,000 homes—while promising OpenAI will foot its own infrastructure bills so residents don't pay the tab.
Demos circulating on X show an unreleased Claude checkpoint dubbed "Fable 5.5" generating animations without any video or image models. In one test, it autonomously wrote Python code for over two hours to produce a standalone HTML animated film. Anthropic has not confirmed the model exists.
Moonshot AI just released the full open weights for Kimi K3.1 Pro, a 2.8-trillion-parameter MoE model that activates 104B parameters per token. It features a 1M token context window and native multimodal reasoning. Its massive 1.56TB size targets enterprise clusters running vLLM or SGLang.
DeepPrinciple just dropped Hydrogen-1.0, an open-weight, 753-billion-parameter model designed for chemistry, materials science, and biology. Fine-tuned on real laboratory workflows, it can write and self-debug simulation code, decipher NMR spectra, and compile protocols directly for lab automation. It even matches or outperforms GPT-6-Astra on several core scientific benchmarks.
Levanto just launched Sage2, an API duo that extracts typed data and evaluates real-time decisions from live conversation. Its streaming engine updates answers on the fly as context grows without resending chat logs, while extraction is claimed to be 166x cheaper than Opus 5.5. Why it matters: agent builders can now handle mid-chat mind changes and dynamic routing with minimal latency.
Researchers introduced Weight-Space Programming, showing OpenAI's GPT-6 Astra can natively interpret and tweak neural network weights without traditional gradient training. By altering just 0.016% of Qwen3.5-0.8B's weights, it fixed a reasoning loop and bumped math accuracy from 7.9% to 48.1%. Why it matters: Direct parameter steering hints at automated model patching, though current results rely heavily on external Python tools rather than pure weight intuition.
Anthropic just launched Message Threads in public beta for the Claude Messages API. Long-running agents no longer have to re-upload full conversation histories on every turn, sending only new messages instead. For heavy sessions with 500k to 1M tokens, median time to first token drops by roughly 35%.
Runway just launched Dev plugins for Claude Code, Codex, and Cursor, connecting coding agents directly to its generative media platform via MCP. Agents can check credit balances, pick models, and trigger video workflows right inside the editor without storing API keys. It basically turns your coding agent into an end-to-end multimedia builder.
xAI just introduced Data Bot in Grok Bot, an internal AI teammate that lets any employee query data warehouses and run deep analyses without waiting in line. Curated by data scientists with shared institutional memory, it can live in Slack to post updates and handled an estimated 70 hours of manual work in just one week.
Anthropic just open-sourced FlashProve, a cryptographic system using zkSNARKs to prove an AI response genuinely came from a specific model without revealing its private weights. Running on a single H100 GPU, it generates proofs for a 400B-parameter Llama 4 Maverick at roughly 110 tokens per second, which can be verified on a standard CPU in seconds. While proof generation is 70–95x slower than standard inference, it creates a practical defense against rogue model substitutions and tampering.
Andrej Karpathy just built HackerTube, an alternative YouTube recommender, in about two hours using Claude Opus 5.5. Frustrated by YouTube's noisy recommendations, he used Claude Code to filter 43 million Hacker News comments down to 6,454 heavily recommended videos with no login required. Why it matters: It proves how frontier AI lets solo developers ship custom, algorithm-free tools in a single afternoon.
StudyOnTV just turned screen time management into a learning gate for Android TVs and tablets, forcing kids to answer questions before earning viewing minutes. It runs completely locally with no cloud accounts, and can plug into AI models like DeepSeek to auto-grade subjective answers.
OpenAI just revealed that starting in December, ChatGPT and Codex will share a single token allowance pool across Plus, Pro, and Business plans. The shift comes as OpenAI brings Work's agentic tools directly into regular chat, meaning heavier models and complex tasks will draw from the same cap, though instant chat stays untouched. If you burn through your quota, you will have to wait for resets or buy extra credits.
DeepSeek just open-sourced dsh-superagent-prototype, an experimental background agent designed to handle GitHub tasks without cluttering human attention. It runs tasks in isolated sessions and only pings developers when a pull request is ready or human intervention is strictly required. Keep in mind that DeepSeek frames this as research code rather than an official product.
xAI just announced that Grok API text outputs will include invisible watermarks by default starting November 7, 2026, to comply with the EU AI Act. The system subtly tweaks token selection during generation without adding extra characters, and admins can opt out per model in the console.
