AI Daily Digest · July 5, 2026
1. Global AI Investment Hits Record $510B in H1 2026 — AI Took 70%+ of Q2 Capital
Crunchbase's H1 2026 global venture report, published July 2, reveals an extraordinary milestone: global startup investment hit $510 billion in the first six months of 2026 — more than all of 2025 combined. The previous half-year record was $375 billion (H2 2021). This one topped it by $135 billion.
The AI concentration is staggering. Over 70% of Q2 global startup capital went to AI-focused companies, up from roughly 50% a year earlier. OpenAI and Anthropic alone raised $217 billion — 43% of all H1 venture funding. OpenAI's $122 billion Q1 round was the largest private funding in history, while Anthropic's $65 billion Series H in late May pushed its valuation to $965 billion. Sixteen companies raised billion-dollar rounds in Q2, totaling $108.6B — seven of them were frontier AI labs.
The exit market is equally dramatic. Thirty-two venture-backed companies went public above $1 billion valuation in Q2. Twenty-four more were acquired at $1B+, totaling $113 billion in M&A — the highest single-quarter acquisition total on record. If Q3 comes in anywhere near Q2's $205 billion pace, 2026 will become the first calendar year to cross $1 trillion in venture investment.
— Crunchbase
🔗 Crunchbase: Global Startup Investment Hit Record $510B In H1 2026 · AI IPO Tracker 2026 · Anthropic $65B Series H — TechCrunch
2. Anthropic Launches Claude Sonnet 5 — The Most Agentic Sonnet Yet
Anthropic released Claude Sonnet 5 on June 30, calling it the most agentic Sonnet model to date. It can make plans, use tools like browsers and terminals, and run autonomously at a level that previously required Opus-class models. Performance is close to Opus 4.8 across reasoning, tool use, coding, and knowledge work, but at significantly lower prices.
Early access partners reported dramatic improvements. Testers described how Sonnet 5 finishes complex multi-step tasks where Sonnet 4.6 would stall, autonomously checks its own output, and handles sustained coding, tool use, and debugging across messy codebases. One Rust engineer reported asking Sonnet 5 to investigate a bug — it unpromptedly wrote a reproducing test, implemented the fix, then stashed the change to confirm the bug returned without it, all in a single pass.
Sonnet 5 launches with introductory pricing of $2/M input tokens and $10/M output tokens through August 31 (then moving to $3/$15). It replaces Sonnet 4.6 as the default model across all plans. Safety evaluations show lower rates of hallucination and sycophancy than Sonnet 4.6, with real-time cyber safeguards enabled by default.
— Anthropic
🔗 Claude Sonnet 5 Announcement — Anthropic · Claude Sonnet 5 System Card · Sonnet 5: Near-Opus Agentic Coding at Mid-Tier Prices — Geek Source Codes
3. Kling AI Raises $2.8 Billion at $18 Billion Valuation
Kling AI, the generative AI video platform spun out from Kuaishou, raised over $2.8 billion in a funding round led by Alibaba and Tencent, pushing its post-money valuation to $18 billion — the largest funding round ever completed by an AI video model developer globally. Baidu also participated alongside financial institutions including BlueFive Capital, CPE Yuanfeng, and CITIC Securities.
The timing is strategic. OpenAI shut down the consumer version of Sora in April 2026 due to high computing costs and low monetization. Runway also shifted focus away from commercial consumer video tools. This leaves the global video AI market wide open, and Kling is stepping directly into the void with its Kling 3.0 platform, which generates 4K video clips from text prompts.
Kling AI is already generating real revenue — first quarter 2026 revenue exceeded $96 million (650 million yuan), a 300% increase year-over-year, with annualized revenue approaching $500 million. Kuaishou, which will dilute its ownership to ~68%, plans to list Kling AI on the Hong Kong Stock Exchange by mid-2027.
— VentureBurn · Kuaishou HKEX Filing
🔗 Kling AI Raises $2.8B at $18B Valuation — VentureBurn · Kuaishou HKEX Filing
4. Z.ai Launches ZCode — A Free AI Coding IDE Powered by GLM-5.2
Z.ai (Zhipu AI) launched ZCode, a free desktop "Agentic Development Environment" (ADE) purpose-built for its GLM-5.2 model, available on macOS, Windows, and Linux. The tool challenges established players like Cursor, Claude Code, and GitHub Copilot with an agent-first architecture where users describe an outcome and the agent plans, edits files, runs checks, and iterates until the goal is met.
ZCode's distinctive features include multi-device continuity — users can steer a running coding agent from WeChat, Feishu, or Telegram on a phone — and a BYOK (Bring Your Own Key) architecture that supports third-party models and self-hosting. Teams can download GLM-5.2 (MIT-licensed open weights, 744B parameters MoE, 40B active), host it on their own infrastructure, and run ZCode without touching Z.ai's cloud, eliminating both US export-control risk and Chinese data-sovereignty concerns.
The pricing undercuts Western competitors significantly — Lite plan at $16.20/month and Max at $144.00/month. GLM-5.2 itself ranks #2 on Code Arena globally (trailing only Claude Fable 5) at API costs 82% lower than Claude Opus 4.8 ($1.40 vs $5.00 per million input tokens). The launch comes amid heightened awareness of "sovereign access risk" following the June 12 US export control directive that temporarily suspended access to Anthropic's Fable 5 for foreign nationals.
— VentureBeat · Z.ai
🔗 Z.ai Launches ZCode — VentureBeat · GLM-5.2 on Hugging Face · AI Coding Agents 2026 Comparison — LushBinary
5. Claude Science — Anthropic Enters the Drug Discovery Race
Anthropic shipped Claude Science on June 30, an AI agent platform designed to automate biological and chemical research workflows. Claude Science integrates 60+ scientific databases across genomics, proteomics, structural biology, and cheminformatics, connecting to NVIDIA BioNeMo and Basecamp Research's EDEN dataset. It handles the full pipeline from literature search to manuscript review, with a dedicated reviewer agent checking citations, calculations, and figure accuracy.
The talent signal is even bigger than the product. John Jumper — who shared the 2024 Nobel Prize in Chemistry for AlphaFold and spent nine years at Google DeepMind — joined Anthropic on June 19. His presence changes how pharmaceutical companies and government research funders assess Anthropic as a partner. The move completes a three-way race between Google DeepMind (AlphaFold/Isomorphic Labs), OpenAI (GPT-Rosalind, launched April), and Anthropic for control of AI-driven drug discovery.
Claude Science is available to Pro, Max, Team, and Enterprise subscribers, with $30K in research credits available (deadline July 15). More than 200 AI-discovered drug candidates are in clinical development — but zero have FDA approval, underscoring the gap between research acceleration and clinical outcomes.
— Anthropic · MIT Technology Review
🔗 Claude Science: An AI Workbench for Scientists — Anthropic · Claude Science Puts Anthropic in Drug Discovery Race — Awesome Agents · AlphaFold Nobel Laureate John Jumper Joins Anthropic — TechTimes
6. NVIDIA Launches Halos for Robotics — First Full-Stack Physical AI Safety System
NVIDIA released Halos for Robotics on June 22, the industry's first full-stack robotics safety system that integrates AI compute and safety capabilities into a unified architecture for robot development, validation, and industrial deployment. As humanoid robots and physical AI move from pilots to mass deployment, isolated safety features are no longer sufficient — Halos provides a standardized architecture covering hardware, software, sensors, applications, and certification.
The system comprises three layers: NVIDIA IGX Thor and Holoscan Sensor Bridge (hardware — industrial-grade AI compute with built-in safety mechanisms), NVIDIA Halos OS (software — including safety-critical runtime functions and external perception safety blueprints that use external cameras and AI agents to dynamically adjust robot behavior), and the Halos AI Systems Verification Lab (the world's first ANAB-accredited program covering both functional safety and AI safety for physical AI, preparing partners for third-party certification with TÜV Rheinland, UL Solutions, and others).
Agility Robotics will be the first to adopt Halos for its industrial humanoid robots deployed in factories, warehouses, and logistics. This marks a critical step toward making humanoid robots safe enough to work alongside humans — a prerequisite for physical AI at scale.
— IT之家 · NVIDIA
🔗 NVIDIA Halos for Robotics — IT之家 · The Quest to Make Humanoid Robots Safe Enough for Humans — WSJ · Physical AI Ecosystem 2026 — Seeking Alpha
7. Five AI Labs Back a Common Jailbreak Safety Scale
Five major AI labs have reportedly agreed to back a common jailbreak scoring scale ahead of an August 1 standards target, according to reports from July 3-4. The initiative is an early step toward standardized AI model safety testing that would allow regulators and enterprise buyers to compare model security across vendors on a consistent scale.
The development comes amid intensifying safety debates following the GPT-5.6 "High Risk" classification covering all three variants (Sol, Terra, Luna) across cybersecurity and biological/chemical domains. Chain-of-thought spoofing attacks have also emerged as a new security risk for reasoning AI models, with researchers demonstrating that models' internal reasoning chains can be manipulated to bypass safety guardrails.
A standardized jailbreak scale would address a growing pain point for enterprise AI buyers: without consistent safety metrics, procurement teams struggle to compare models from different providers. The August 1 deadline suggests the industry is racing to self-regulate ahead of potential government intervention, particularly as the IPO window for major AI labs opens in late 2026.
— Creati.ai · Multiple Sources
🔗 Five AI Labs Back Common Jailbreak Safety Scale — Creati.ai · Chain-of-Thought Spoofing Security Risk — Creati.ai · GPT-5.6 Safety High Risk — Westenet
Next digest: July 6, 2026
Top comments (0)