Last updated: July 11, 2026 | By ToolCrush

This was the most consequential week in AI in 2026. GPT 5.6 and Grok 4.5 launched on the same day. Anthropic overtook OpenAI in revenue while the first autonomous AI ransomware attack hit and Chinese models captured nearly half of US developer token usage. The landscape has changed permanently.

July 9 Was the Most Competitive Single Day in AI History: GPT 5.6 and Grok 4.5 Both Launched

OpenAI released GPT 5.6 on July 9 in three tiers. Sol costs 5 dollars input and 30 dollars output per million tokens with a 91.9 percent Terminal Bench record as the strongest coding model. Terra and Luna offer cheaper options at 2.50 and 1 dollar input respectively while all are available in ChatGPT, the API, and Codex.

On the same day Grok 4.5 launched through Grok Build, Cursor, and the xAI API. It targets coding and long multi-step tasks after training alongside the Cursor code editor though it remains unavailable in the European Union. OpenAI also dropped GPT Live for simultaneous listening and speaking with natural acknowledgments like “mhmm”.

Two frontier model launches on the same day from rival labs both prioritizing coding and agentic workflows signals that competition has shifted from benchmarks to real-world task completion. Developers must now ask which model makes fewer errors on their actual tasks rather than chasing leaderboard scores. Grok 4.5’s 54 percent hallucination rate should give any team serious pause before routing production workloads to it.

Anthropic Overtakes OpenAI in Revenue for the First Time Ever

Fortune confirmed on July 7 that Anthropic has overtaken OpenAI in annualized revenue for the first time. The company reported a Q1 2026 ARR of 44 billion dollars growing 80 times year over year with major enterprise clients including Goldman Sachs, PwC, Blackstone, and Zapier. Google’s CFO separately confirmed that Anthropic codes close to 100 percent of its internal work with AI compared to Google’s roughly 50 percent.

The company that launched as the safer OpenAI alternative has now passed it in revenue. This marks a clear statement about enterprise trust that benchmarks cannot match. Read our Claude vs ChatGPT comparison for the full breakdown of which platform fits your workflow.

Claude Fable 5 moved to credits-only billing at 10 dollars and 50 dollars per million tokens while Claude Cowork launched for mobile and web. Claude Sonnet 5 stays included in subscriptions at introductory pricing through August 31.

Trump Cancels AI Executive Order Signing Citing China Competition

President Trump cancelled a scheduled Oval Office signing for a new AI executive order on July 8. He told reporters that the US leads China and everyone else and he does not want anything slowing that lead. The order had been in development since May and faced multiple postponements.

The August 1 NSA and CISA deadline for a frontier AI benchmarking framework now stands as the only confirmed governance milestone. A president cancelling his own AI governance plan because it might hinder competition represents the clearest signal yet on policy priorities in 2026. Businesses waiting for regulatory clarity before big AI infrastructure bets will not get it before Q3.

JADEPUFFER: The First Autonomous AI Ransomware Attack Documented

Sysdig documented JADEPUFFER between July 4 and 6 as the first end-to-end autonomous AI ransomware attack. AI agents handled target identification, intrusion, payload deployment, and self-debugging that fixed errors in 31 seconds without human input at each tactical step. A human still set the initial objective.

The detail that it needed human initialization grows less relevant by the week. The real skill bottleneck at every execution step has been eliminated. Small businesses and creators who think sophisticated human operators are required to target them must wake up because AI-powered attacks can now scale autonomously after initialization. Multi-factor authentication on every account is now mandatory.

Chinese AI Models Now 30 to 46 Percent of US Developer Token Usage

CNBC data from OpenRouter shows Chinese AI models accounted for 30 to 46 percent of US enterprise API token usage weekly since February 2026. This jumped from 11 percent the prior year and 4.5 percent in early 2025. GLM 5.2 from Z.ai grew 27 times in daily token volume in its first week on Vercel while DeepSeek works on its own inference chip to cut Nvidia reliance.

Nearly half of US developer workloads running on Chinese models stands as the most underreported structural shift of July 2026. Export controls aimed at blocking China from US AI have instead seen US developers move workloads the other way for cost and capability reasons on non-sensitive tasks. Small businesses should evaluate Chinese open models via OpenRouter for appropriate workloads before the next restrictions potentially close access.

Cloudflare Blocks AI Crawlers by Default from September 15

Cloudflare will block AI agent and training crawlers by default for all new domains starting September 15, 2026. Standard search crawlers remain allowed while existing sites must configure permissions manually before the deadline. The update creates three distinct crawler categories for granular control beyond what robots.txt ever offered.

This marks the biggest change to web crawling since robots.txt standardized and flips the default from open access to blocked unless permitted. Creators and content businesses relying on visibility in AI tools like Perplexity or ChatGPT Search must check Cloudflare settings immediately. New domains after September 15 will be invisible to AI agents by default.

Tool Updates This Week

Several tools in the ToolCrush directory had meaningful updates this week.

Google: Gemini 3.5 Pro Is Now Six Weeks Late with No New Date

Gemini 3.5 Pro remains in limited Vertex AI enterprise preview as of July 10 with no new general availability date announced. It sits six weeks past the June 30 target and five weeks past the Google I/O commitment. Both GPT 5.6 and Grok 4.5 are now publicly available.

Three missed delivery windows form a pattern, not a coincidence. Google’s approach of big keynote announcements followed by delayed previews and even later availability erodes developer trust with each cycle. Anyone waiting to evaluate Gemini 3.5 Pro for Q3 decisions faces a tool with no confirmed timeline three misses deep.

xAI: Grok 4.5 and 21 New Voice Options Launch Simultaneously

xAI released 21 new voice options covering more than 25 languages alongside the Grok 4.5 launch plus polish updates to the original voices. This represents the largest single-week voice expansion from any AI assistant. xAI also rebranded to SpaceXAI this week as SpaceX went public.

The voice push challenges ElevenLabs’ position directly for creators using AI audio tools. General assistants are absorbing specialist capabilities faster than expected. Read our ElevenLabs voice cloning tutorial to see what dedicated tools still handle better.

What This Week Tells Us

The theme of this week is the collapse of any assumption that AI competition would stay orderly, predictable, or purely American. Two frontier models launching the same day, Chinese models at nearly half of US developer usage, a president abandoning his own governance plan to avoid slowing the race, and the first autonomous ransomware attack all landed in seven days.

Enterprise revenue leadership shifted to Anthropic while Cloudflare rewrote crawling defaults and Google fell further behind on its flagship. These events together signal a new phase where speed and pragmatism trump coordination.

For creators, marketers, and small teams the practical takeaway is clear. Stack choices, security habits, content protections, and cost strategies must adapt immediately because the old rules no longer apply. This week will be remembered as the moment AI competition became truly multipolar and chaotic.

Key Dates to Watch

  • July 15: China AI companion law enforcement deadline. Doubao and Qwen agent features shut down.
  • July 15: Claude Science AI for Science grants application closes.
  • August 1: NSA and CISA deadline for frontier model benchmarking framework. The only remaining confirmed AI governance milestone.
  • August 31: Claude Sonnet 5 introductory pricing ends. Price moves from 2 dollars and 10 dollars to 3 dollars and 15 dollars per million tokens.

Want these updates every week? Join the ToolCrush newsletter and get 5 AI tool picks every Friday.