🗞️ AI Daily Briefing — 2026-07-02

🔥 Top Story

Anthropic launches Claude Sonnet 5 and makes it the default for all users — the first mid-tier model to beat its Opus sibling on a major coding benchmark. Released June 30 and rolled out as the default model for every Free and Pro user on July 1, Sonnet 5 scores 80.4% on Terminal-Bench 2.1 vs. Opus 4.8’s 74.6%, while coming within striking distance on SWE-bench Pro (63.2% vs. 69.2%). Introductory pricing is aggressive at $2/$10 per million tokens (input/output) through August 31 — cheaper than Sonnet 4.6. Anthropic is positioning this as the go-to model for agentic workloads, with dramatically improved prompt-injection resistance (attack success drops from 31.5% to 0.93% in browser-use scenarios). (TechCrunch) (DataCamp)

🚀 Model & Research News

  • Fable 5 is back online globally after 19-day export ban. The US Department of Commerce lifted export controls on Anthropic’s Claude Fable 5 and Mythos 5 as of July 1. Anthropic deployed a new classifier that blocks the disputed jailbreak technique in over 99% of cases. Fable 5 is included at up to 50% of weekly usage limits for Pro/Max/Team plans through July 7. (CNBC) (MacRumors)

  • Google drops Nano Banana 2 Lite (Gemini 3.1 Flash-Lite Image) — 4-second image generation at $0.034 per 1,000 images. Launched June 30, this is Google’s fastest and cheapest image generation model, optimized for high-throughput enterprise workflows rather than artistic quality. Available via Google AI Studio and the Gemini API. (VentureBeat)

  • xAI rolls out Grok 4.1 to all users. Available now on grok.com, X, and mobile apps in Auto mode. xAI also introduced /goal in Grok Build — a long-running autonomous mode for larger implementation tasks with plan, execute, pause, and resume controls. (xAI)

  • Google Search gets its biggest upgrade in 25 years — AI agents powered by Gemini 3.5 Flash. AI Mode now has 1B+ monthly users globally. Users can create, customize, and manage AI “information agents” that monitor topics 24/7 and take actions like tracking flights or organizing emails. (Google Blog)

🛠️ Tools & Developer Updates

  • dbt Labs: AI-driven acceleration is outpacing trust and governance. The latest State of Analytics Engineering Report shows trust in data rose from 66% to 83% YoY, but organizations are struggling to keep governance frameworks in step with AI adoption speed. Virtual event on July 22 to demo dbt Wizard in the CLI. (dbt Labs)

  • LlamaIndex expands into agentic OCR and document workflows. The framework is pushing beyond pure RAG with ParseBench, LiteParse, and Visual Document Retrieval — targeting the unsexy-but-massive document parsing problem that blocks most enterprise RAG deployments. (LlamaIndex)

  • PostHog teases “What if your product built itself?” feature. Details sparse, but the product analytics platform published a product update on June 22 hinting at AI-powered product development capabilities. Virtual event on July 15 covering warehouse compute savings. (PostHog)

💰 Funding & Business

  • National Grid invests $1.75B for 35% stake in Joulent — AI power infrastructure becomes institutional. The UK energy giant’s first major move into US AI data center power, backing a 2.67 GW gas-fired plant in West Texas (Project Kilby) that will supply a Microsoft-operated data center under a 20-year PPA. First power delivery targeted for 2028. The deal signals that AI’s energy appetite is now big enough to attract utility-scale capital. (Bloomberg) (Yahoo Finance)

  • AI drives global equity issuance to $729.4B in H1 2026 — up 73% YoY. AI-linked companies accounted for a massive share of the surge, led by SpaceX’s record IPO. Anthropic’s IPO filing and OpenAI’s approaching public listing are the next catalysts. (Yahoo Finance)

  • Applied Digital delivers 75 MW of operational AI capacity at Polaris Forge 1. Phase 1 of Building 2 is now Ready for Service, adding to the company’s growing AI data center footprint. (Applied Digital)

🐦 Notable from the Timeline

  • John Jumper (AlphaFold creator) confirmed leaving Google DeepMind for Anthropic after nearly nine years. This follows Noam Shazeer’s departure to OpenAI — Google is hemorrhaging its most senior AI talent as pre-IPO equity at Anthropic and OpenAI proves impossible to match. (CNBC)

  • Google DeepMind lost 6 senior researchers to Meta, OpenAI, and Anthropic in 5 months after its internal “AI Coding Strike Team” pivot. The brain drain is accelerating as the lab tries to close Gemini’s gap in agentic coding. (TechTimes)

  • White House executive order (June 2) now entering implementation phase — AI developers will voluntarily give the government early access to frontier models for up to 30 days before release. An “AI cybersecurity clearinghouse” is being formed by Treasury, NSA, and CISA. (White House)

  • UN scientific panel warns AI development is outpacing scientific understanding and government oversight. The UN and ITU launched the AI for Good Global Commission to address the gap. (JD Supra)

📊 Benchmark Watch

The Arena (formerly LMArena) leaderboard is the tightest it’s ever been — the top 5 models are clustered within ~55 ELO points. Claude Opus 4.8 holds #1 at ~1510+ ELO (also tops the coding leaderboard). GPT-5.5 Pro, Gemini 3.1 Pro Preview, and Claude Opus 4.7 follow close behind. Fable 5 briefly topped the board before its June 12 suspension — now that it’s back, expect a shakeup this week. Sonnet 5 hasn’t been fully ranked yet but its Terminal-Bench 2.1 score suggests it’ll punch above its weight class. (Arena.ai)

🎙️ Podcast Highlights

  • TBPN (now OpenAI-owned since April) continues daily coverage of AI developments. The show is on track for $30M in ad revenue this year, up from $5M in 2025. No standout guest clips from the last 24h, but the OpenAI acquisition itself remains one of the more interesting media-AI convergence stories of the year. (CNBC)

🔗 Worth Reading