According to Google Research: Google Research reports recent results from Antigravity's teamwork multiagent systems combined with Gemini 4 models, which tackle complex problems across mathematics, computer science, and engineering. According to Google, these tools pair parallel generation with dedicated verifiers to bridge the gap in model predictability and partition context to maintain multi-day task accuracy. Google is making these agentic tools available externally to researchers and developers.
OpenAI is launching ChatGPT Wallet for its US Pro subscribers, allowing its Dots AI agents to make online purchases. The system uses biometric verification for every transaction and generates payment tokens so your actual credit card details are never shared. In the future, OpenAI plans to let users fully delegate purchases by setting strict limits and pre-approval rules.
Google has announced Gemini 4 Hydrogen, a new AI model focused on speed and efficiency for everyday coding and agent tasks. According to Google, it is 2.5 times faster than the previous generation model and is already rolling out globally, including for free users.
Researchers have developed LLMON, an LLM-native markup language designed to explicitly communicate structure and semantics to models, bypassing the limitations of traditional JSON. By separating instructions from data using reserved special tokens, the newly released experimental Granite 4.1 LLMON preview models achieve significantly higher resistance to prompt injection attacks while maintaining general capabilities.
According to Google DeepMind: Google is introducing Gemini 4 Hydrogen, a new model designed for rapid execution and everyday agentic workflows. Available across Google AI Studio, the Gemini API, and the Gemini app, Hydrogen offers flat pricing at $1.50 per million input tokens and $7.50 per million output tokens while delivering significant speed and efficiency upgrades over its predecessor.
According to Google AI: Google is rolling out Gemini 4 Hydrogen to Search globally, marking its biggest model upgrade to date. The update brings better intent understanding for complex questions, improved multi-topic query linking, more visual response formats, and richer language capabilities for everyday users.
Chinese AI company MiniMax has open-sourced its flagship image generation model, H3-Image, featuring native 4K resolution and advanced spatial reasoning. The model aims to eliminate the typical artificial look of AI images by handling intricate details, complex text layouts, and even CAD-style mechanical breakdowns. Positioned as a strong competitor in multimodal AI, the model is available via open-source and low-cost API pricing.
Meta has rolled out dedicated @muse.ai email addresses for its personal AI agent, Muse. Users can now assign the AI to handle tasks such as sending reservation emails or summarizing long messages, with availability expanding from paid subscribers to all users over time.
A roundup of major tech news: ByteDance is developing an AI assistant similar to Meta Muse, while Anthropic enforces strict regional bans on Claude. Meanwhile, OpenAI reports strong third-quarter revenue reaching $10 billion, and Apple announces the permanent closure of its support community.
According to LlamaIndex Blog: LlamaParse has released ParseBench-2, a new document parsing benchmark that introduces human-defined severity scores to better evaluate parser errors. The updated benchmark more than doubles its document corpus, adds forms as a sixth dimension, and ranks 41 default parser configurations, with LlamaParse Agentic Plus leading overall.
Anthropic's top-tier AI model Claude Fable 5.2 was tested against Opus 5.5 and Sonnet 5.5 on tasks like summarizing lengthy government reports and preparing presentation slides. While Fable 5.2 excels at understanding user intent and crafting user-friendly slides, it is significantly more expensive and time-consuming, making Opus and Sonnet more practical choices for everyday tasks.
According to Google AI: Google has introduced the Gemini Decisions API, powered by a new Gemini Decision 1 model. It is built for fast, bounded choices like tool selection and policy checks, processing multimodal inputs and up to 1,000 questions in a single forward pass without generating answer text token by token. This slashes end-to-end latency to 150-250 ms and cuts costs significantly.
According to Google DeepMind: Google has announced AlphaGenome 2, an AI model that predicts how DNA sequence variations impact specific human cell types and disease states. Trained on over 100 million individual cell profiles, the model identifies twice as many causal variants as its predecessor, helping scientists uncover the cellular roots of various diseases.
Databricks has launched an upgraded version of its AI agent Genie, designed to handle complex, multi-step business tasks completely on its own. CEO Ali Ghodsi demonstrated how he uses the tool internally to get unfiltered feedback, such as analyzing employee survey complaints and reviewing failed sales pitches. This unvarnished truth helps leaders overcome communication barriers in large organizations where subordinates often hide bad news from top management.
Baidu just dropped its new Ernie 5.5 Flash Preview model, focusing heavily on advanced programming capabilities. Beyond writing simple code, this model can independently handle complex end-to-end tasks like generating 3D environments, building playable games, and editing videos using automated loops, matching top-tier global competitors.
Tether AI Research introduces BotPsy, a 0.8B-parameter model designed for on-device agentic workflows. Built on Qwen3.5-0.8B, it achieves state-of-the-art tool use for its size, outperforming many larger models while keeping user data private.
Gemini Notebook lets companies build custom AI helpers by simply uploading existing manuals and documents. It answers routine employee questions with citations and generates FAQs or handover materials instantly, saving countless hours for HR and administrative teams.
According to AWS Machine Learning: Moonshot AI's Kimi K3.1 Pro Highspeed is now available on Amazon Bedrock, bringing 2 to 3 times faster output generation. It retains the powerful 1-million-token context window while adding major boosts for complex coding, financial modeling, and agentic workflows.
According to Google DeepMind: Google has officially released Gemini 4 Carbon, its most powerful AI model designed for complex coding, scientific research, and cybersecurity tasks. It builds on the earlier Argon model by offering deeper reasoning, multi-agent workflow support, and top-tier performance on industry benchmarks at a low cost.
Recent studies show that ChatGPT commands over 81 percent of the global AI chatbot traffic, leaving competitors far behind. Meanwhile, Anthropic launched its new Claude 5.2 models, boasting a twenty percent drop in API costs and a thirty percent boost in generation speed.
According to Meta AI Research: Meta has launched Muse Ember, a lightweight AI model built specifically for repetitive agent tasks like tool calls, coding, and document analysis. It delivers high intelligence at a fraction of the cost, rivaling top models while priced at just $0.10 per million input tokens.
Large AI models use a "Mixture-of-Experts" design to save power, but most experts sit idle while taking up massive memory. Traditional pruning methods cut the weakest individual experts, which often breaks cooperative groups. A new technique called HOPE accounts for how experts work together, allowing researchers to trim up to 50% of a model's experts while maintaining high performance, especially in complex coding tasks.
According to Anthropic News: Anthropic has upgraded its heavy-duty AI models to versions 5.2, bringing massive improvements in coding, reasoning, and scientific problem-solving. The new models are faster, cheaper, and capable of complex tasks like writing operating systems from scratch, solving black hole physics problems, and formalizing mathematical proofs.…
According to Google AI: Google has launched Gemini Omni 1.2 Flash, its updated AI video generation model that now supports 15-second clips and costs 30% less than the previous version. The new model offers better reference consistency, conversational editing, and is now generally available across developer and enterprise platforms.
According to Thinking Machines Lab: Inkling 2 Small is a new Mixture-of-Experts AI model with 594B total parameters, offering massive upgrades in coding, vision, and reasoning. Initially available with closed weights on Tinker, its full Apache 2.0 weights will be released next week following safety checks.
According to World Labs: Researchers have released Frameboy, a new demo that lets users explore 3D worlds generated by the Atlas world model in real-time at up to 15 fps. By streaming frames from cloud GPUs straight to the browser, the tool removes the frustrating wait times of traditional AI video generation. This speed lets creators explore and iterate instantly, completely changing how people interact with generative 3D spaces.
