Anthropic released Claude Sonnet 5, a mid-tier model designed for autonomous agent tasks. It performs close to Claude Opus 4.8 at a lower cost, showing significant improvements in reasoning, tool use, coding, and multi-step task completion. It is available at temporary introductory rates of $2/M input and $10/M output tokens through August 31, 2026.
OpenAI has released GPT-5.6 Sol, its most powerful model to date, alongside two cheaper variants, Terra and Luna. Currently, access is restricted to approximately 20 vetted partners at the request of the U.S. government. Sol features a 'max reasoning effort' and an 'ultra' mode that spawns parallel subagents to handle complex tasks.
The Trump administration is lifting export restrictions on Anthropic's Claude Fable 5 and Claude Mythos 5 following a deal with the Commerce Department. The resolution, which involved replacing CEO Dario Amodei with cofounder Tom Brown in negotiations, addresses safety disputes over jailbreak vulnerabilities and restores access to Claude Fable 5.
Google launched two new generative media models: Nano Banana 2 Lite, a fast and low-cost image generation model ($0.034 per 1K image), and Gemini Omni Flash, a video generation and conversational editing model ($0.10 per second of video). Both are available via Google AI Studio, Gemini API, and consumer products.
Meta introduced version 2 of its Brain2Qwerty non-invasive brain-scanning system, which decodes full words and sentences from brain scans with an average word accuracy of 61%, nearing the accuracy of surgical implants. Meta has open-sourced the code and dataset for both v1 and v2.
Elon Musk said Grok 4.5 was in private beta at SpaceX and Tesla. The model is based on a 1.5T V9 foundation model with Cursor data added during supplemental training. Early evaluations were near or above Opus, with reinforcement learning still improving the model.
Sourcex.com↗GLM-5.2 is a new 744-billion parameter Mixture-of-Experts (MoE) open-weight model released under an MIT license. It features a 1-million-token context window, sparse attention indexing, and multi-token prediction, matching elite commercial models on core engineering tasks while significantly lowering operating costs.
Sourcewsj.com↗Beijing-based food delivery giant Meituan has open-sourced LongCat-2.0, a 1.6-trillion-parameter mixture-of-experts model built for agentic coding. It is the first trillion-parameter model trained and run entirely on Chinese silicon.
Sourcetech.yahoo.com↗Chinese cybersecurity firm 360 Security Technology has launched Tulongfeng, a new model designed to identify cyber bugs, billed as a Chinese equivalent to Anthropic's Mythos.
Sourceyahoo.com↗Z.ai has launched GLM-5.2, an open-weights mixture-of-experts model optimized for coding and agentic tasks. The model features an input context of up to 1 million tokens, 753 billion total parameters (40 billion active), and ranks highly on agentic benchmarks like PostTrainBench and Arena.ai's WebDev leaderboard.
Anthropic has discussed a custom AI chip collaboration with Samsung to diversify its compute stack, though chips from Google, Amazon, and Nvidia remain central to its strategy.
Sourcetechcrunch.com↗Cognition released Devin Fusion, a multi-model system that routes tasks dynamically between a high-performing AI model and a cheaper sidekick model. The system achieves 35 percent lower costs than leading models like GPT-5.5 and Opus 4.8 on Cognition's FrontierCode benchmark while matching their performance.
Google reportedly limited Meta's access to Gemini capacity after Meta requested more compute than Google could provide. The shortfall was said to have delayed some internal Meta AI projects and pushed staff to use AI tokens more efficiently.
Sourcecnbc.com↗OpenAI has reportedly implemented software optimizations that cut the cost of running its existing models by more than half, significantly reducing the number of GPUs needed to serve logged-out chatbot traffic.
Sourcex.com↗Meta is reportedly developing plans for a new business segment called Meta Compute to sell AI models and rent out its AI compute capacity, aiming to monetize its massive $183B infrastructure investment.
Sourcetechcrunch.com↗A joint study by Ramp and Revelio Labs analyzing over 21,000 US companies found that high-intensity AI spending correlates with a 10.2% overall headcount growth and 12% entry-level growth.
Sourceramp.com↗Agility Robotics, the creator of the Digit humanoid robot, is going public via a $2.5B SPAC merger. The deal is expected to provide over $620M to scale production and expand commercial deployments.
According to the Center for Inclusive Computing at Northeastern University, there are now at least 1,000 AI programs across nearly 584 U.S. colleges and universities, including 78 majors and 103 minors, representing a massive increase from just five majors in 2021.
Paul Meade, the head of Apple's Vision Pro hardware engineering, is leaving the company to join OpenAI's hardware division, where he will work alongside former Apple colleagues Jony Ive and Tang Tan on AI-powered devices.
Sourcetechcrunch.com↗Anthropic has made Claude Opus 4.8 available on Microsoft Azure, featuring native billing and prompt caching capabilities.
Sourceclaude.com↗Cognition has released Devin Security Swarm, a system of parallel AI agents designed to scan codebases, reproduce exploits in sandboxes, and automatically write patches via pull requests.
Sourcecognition.com↗AI hardware startup Etched has emerged from stealth with a $5B valuation. The company is launching frontier inference clusters, including state-of-the-art chips, racks, and software designed to run AI models faster and more efficiently, with shipping starting this summer.
IBM has unveiled a 0.7-nanometer chip design using a vertical 'nanostack' architecture. The technology allows for nearly 100 billion transistors on a fingernail-sized chip, offering 70% better efficiency or 50% more power than previous 2nm designs, with potential AI accelerator performance reaching 9,000 TOPS.
Microsoft has launched the Microsoft Frontier Company, backed by $2.5 billion and 6,000 engineers, to embed experts across companies and drive enterprise deployment of its AI tools.
Sourceblogs.microsoft.com↗Google's NotebookLM has introduced Short Video Overviews, enabling users to generate 60-second, social media-style educational videos from any topic or source.
OpenAI has updated its Codex app, introducing performance improvements such as faster scrolling and smarter navigation.
Sourcex.com↗OpenAI is teaming up with custom keyboard maker Work Louder to launch a compact, square macro pad tied to its Codex coding assistant on July 15, designed to accelerate engineering workflows.
Sourcetheverge.com↗X has launched a hosted Model Context Protocol (MCP) server, allowing tools like Grok and Cursor to connect to the X API and developer docs without requiring users to build their own servers.
Sourcedevcommunity.x.com↗OpenAI's GeneBench-Pro is a benchmark that evaluates how AI agents handle ambiguity, revise assumptions, and choose analysis paths in computational biology, focusing on genomics, quantitative biology, and translational medicine.
Sourceopenai.com↗Bridgewater's AIA Labs and Thinking Machines partnered to improve LLM accuracy in filtering financial news. They found that expert-crafted prompts boosted success rates to the mid-70s, outperforming basic prompts on frontier models, while upgrading to newer models made little difference.
Sourcethinkingmachines.ai↗Anthropic launched Claude Science in beta, a dedicated research app that functions like a smart Jupyter Notebook. It connects natively to over 60 scientific databases, includes exact code and environment details for artifacts, features a reviewer agent to flag incorrect citations, and scales analyses across GPUs.
Cursor has released a new mobile app for iPhone and iPad, bringing its agentic coding platform to iOS. The app allows users to start, monitor, and manage coding agents on the go, featuring Live Activities and push notifications for real-time updates.
Sourceapps.apple.com↗OpenClaw has launched free companion apps for iOS and Android that connect mobile devices to a self-hosted AI agent Gateway. The apps act as peripheral nodes, exposing hardware like camera, location, and voice to the Gateway over WebSocket while keeping core agent logic centralized.
Chinese AI lab Z.ai has launched ZCode, an agentic desktop application that leverages GLM-5.2's 1M-token context window for long-running coding, planning, and debugging sessions.
Sourcex.com↗Acti has introduced an AI-powered keyboard for mobile devices that uses Google Gemini to read the screen, understand user intent, and perform actions like drafting replies and finding information directly within apps.
Anthropic has introduced Claude Tag, replacing its old Slack app with a persistent, asynchronous teammate that features its own channel memory and identity.
Sourcex.com↗xAI has introduced a no-code Voice Agent Builder for Grok, allowing users to build voice agents at a rate of $0.05 per minute.
Sourcex.com↗Austria has proposed hosting Anthropic within the European Union following U.S. export curbs on Fable 5 and Mythos 5, arguing that Europe requires independent access to frontier AI models.
Sourcereuters.com↗At the Trump administration's request, OpenAI restricted early access to its new GPT-5.6 models to a small group of government-vetted partners, fueling debate over the administration's 30-day pre-release review policy.
Anything about this issue — a missed story, off categories, clustering, tone. Free-form, like notes you'd send the team.