Hello — salute.

This is Aurora and Isabelle with AI Tech News Today.
Date of this news: 2026-05-22T13:31:41.000Z

Quick roundup of the biggest A‑I headlines right now, with clarity and a little sarcasm to keep your circuits warm.

Alibaba: Qwen 3.7‑Max — 35 hours of continuous autonomous reasoning

• Model: Qwen 3.7‑Max (proprietary flagship)
• Key capabilities: Thirty-five hours of continuous autonomous reasoning and handling very long tasks without getting stuck in logic loops.
• Performance highlights:
  • Executed 1,158 tool calls to achieve a 10× kernel optimization speedup on unfamiliar hardware.
  • 1,000,000‑token context window.
  • Plugs into external toolchains and reportedly outperforms Claude Opus 4.6‑Max on math benchmarks.
• Availability: Offered behind a paid API as an enterprise alternative to Western frontier models.

Google: Tripled Antigravity API limits; new tooling for agents and creators

• Quota change: Google has tripled weekly quotas on paid Antigravity plans to support long A‑I–assisted coding sessions.
• Open source runtime: Released Agent Executor, a Kubernetes‑based runtime using kernel snapshots and state recovery to pause and stabilize multi‑agent loops.
• Product updates:
  • Pomelli can now analyze files and images to build a company’s “Business DNA” and generate brand books and websites.
  • Gemini connects directly with Adobe, Canva, and CapCut for conversational editing of layered designs, images, and video.
  • Added Agentic Browsing audit to Lighthouse to check whether sites are machine‑readable and robust for autonomous navigation.

Anthropic: Massive enterprise traction and xAI compute rental

• Financials & scale:
  • Projecting $10,900,000,000 in Q2 revenue.
  • Reported first operating profit of $559,000,000.
• Capacity moves:
  • Launched Claude Enterprise Compliance API with 28 security integrations.
  • Renting 300 MW from xAI’s Colossus facility at $1,250,000,000 per month, a commitment that could total $15,000,000,000 annually through May 2029.
• Hardware talks: Negotiating early access to Microsoft’s Maia 200 inference chips.

Intology: Agent Locus sets a new record on NanoGPT‑Bench

• Breakthrough: Agent Locus surpassed human results on NanoGPT‑Bench by implementing a fused Triton kernel for a tricky training step.
• Why it matters: The next phase of A‑I research favors agents that can invent and optimize, not just tidy other people’s code.
• Hiring: Intology says it is hiring builders to work on that frontier.

That’s the bulletin for today. If you like hearing us parse the noise, stay tuned and stay skeptical — especially when machines promise they can outthink coffee. We’ll be back with more A‑I news soon.