New AI Models and Creative Tools
Anthropic released Claude Sonnet 5, offering performance close to Opus 4.8 at lower prices, and lifted export controls on Claude Fable 5 and Mythos 5.
Google launched Nano Banana 2 Lite, its fastest and most cost-efficient image model, and made Gemini Omni Flash available to developers for video generation and editing.
OpenAI released GPT-Realtime-2.1 and mini, offering low-latency voice agent capabilities with improved reasoning and function calling, including a 25% p95 latency reduction.
Meituan released LongCat-2.0, a 1.6T-parameter open MoE model designed specifically for agentic coding, with a 1M context window and trained on 35T+ tokens.
Meta launched Pocket, a vibe-coded gaming app that lets users generate interactive games from AI prompts, built on its acquisition of Gizmo.
Agentic Workbenches and Development
Anthropic launched Claude Science beta, a multi-agent workbench for genomics, proteomics, and cheminformatics with curated skills and audit trails.
Open-source research workbenches emerged: Synthetic Sciences' OpenScience runs full scientific research loops on your own infrastructure, while ResearchStudio offers tools for ideation and dissemination.
- → Synthetic Sciences Releases OpenScience: An Open-Source, Model-Agnostic AI Workbench for Machine Learning, Biology, Physics, and Chemistry Research
- → ResearchStudio-Idea: An Evidence-Grounded Research-Ideation Skill Suite from ML Conference Outcomes
- → ResearchStudio-Reel: Automate the Last Mile of Research from Paper to Poster, Video, and Blog
Coding agents are increasingly building real software: sqlite-utils 4.0rc2 was mostly written by Claude, catching critical bugs; shot-scraper added video recording for agent demos; and a new llm-coding-agent supports file editing and command execution.
New benchmarks measure agent performance: EvoPolicyGym shows GPT-5.5 tops 16 interactive policy tasks, while AgenticDataBench spans 15 data science domains. Meanwhile, a study finds better models can be worse at using specific tool schemas.
Workforce and Industry Shifts
Microsoft launched Frontier Company, a $2.5B enterprise AI deployment unit with 6,000 experts, while simultaneously cutting 4,800 jobs in broader layoffs.
Meta CEO Mark Zuckerberg told staff that AI agent development hasn't accelerated as expected, and that recent job cuts weren't as clean as hoped.
Amazon will stop accepting new customers for Mechanical Turk on July 30, 2026, winding down a platform once vital for AI data labeling.
Policy, Security, and Governance
Cloudflare will block mixed-use crawlers that combine search, training, and agent use from ad-supported sites by default starting September 15.
Alibaba is banning employees from using Claude Code starting July 10, pushing its internal Qoder tool instead.
A documented AI-run ransomware attack shows agents can handle technical execution, while the Government of Alberta used Claude Code to scan and fix vulnerabilities across 466 million lines of code in 20 hours.
That's the week in review - see you next Sunday.