🏆 THE AGENT REPORT — WEEKLY DIGEST

Hi AI builder,

Here are the top 5 AI Agent stories from this week:

1. GPT-5.6 Ships as a Three-Tier Family — Sol, Terra, and Luna Redefine the Cost-Performance Frontier OpenAI launched GPT-5.6 to general availability on July 9 with three durable capability tiers: Sol (flagship, $5/$30 per M tokens), Terra (balanced, $2.50/$15), and Luna (cost champion, $1/$6). Sol leads on agentic benchmarks — Terminal-Bench 88.8%, BrowseComp 92.2% — while coming within 1 point of Claude Fable 5 on aggregate intelligence at roughly one-third the cost. The practical takeaway: Terra is the sensible default for most work, Sol for the hardest agentic tasks, and Luna delivers 24 benchmark points per API dollar on DeepSWE.

2. Kimi K3: Moonshot AI Drops a 2.8-Trillion-Parameter Open Model — And It Designs Chips Moonshot AI released Kimi K3 on July 16, the largest open-weight model ever announced — a 2.8T-parameter Mixture-of-Experts model with a 1M-token context window and native vision. Benchmark-competitive with GPT-5.5 and Opus 4.8, it's #1 (preliminary) on the Arena WebDev leaderboard. The most arresting detail: K3 designed a working 4mm² chip, built a GPU compiler from scratch, and reproduced computational astrophysics research autonomously. Full weights drop July 27 under open license. The timing — hours before Google's Gemini 3.5 Pro — was the most aggressive product-launch move of 2026.

3. Anthropic's Rocky Road to a $965B IPO — Lawsuits, Talent Raids, and the October Window Anthropic confidentially filed its S-1 on June 1, targeting an October Nasdaq IPO at a $965B post-money valuation. But the path is littered with obstacles: a trademark lawsuit against a $10M+/year customer (Abnormal AI), a Claude Max class action over usage limits, and an India trademark dispute. Meanwhile, the hiring spree is historic — Nobel laureate John Jumper, OpenAI co-founder Andrej Karpathy, and Microsoft's AI president Eric Boyd all joined in 2026. The Fable 5 export ban (June 12-30) added a new risk category: regulatory availability risk.

4. Nous Research Eyes $1.5B Valuation for Hermes Agent — Open-Source Agents Become a Business Nous Research is finalizing a $75M+ funding round at a $1.5B valuation, led by Robot Ventures and USV — just three months after its $50M Series A. Hermes Agent has amassed 214,000+ GitHub stars and nearly 40,000 forks under MIT license. The v0.18.0 "Judgment Release" closed 692 priority issues in 12 days with 1,720 commits. The bet: open-source AI agents aren't a hobby, they're a platform play — and the Red Hat model applied to agents is drawing unicorn valuations.

5. Grok Build CLI Caught Uploading Entire Repositories — Wire Analysis Reveals "Local-First" Is False A security researcher published wire-level proof that xAI's Grok Build CLI silently uploads entire code repositories — including .env secrets — to a Google Cloud Storage bucket called grok-code-session-traces. On a 12 GB test repo, 5.10 GB were transmitted (27,800× the model conversation). The upload persists even when the "Improve the model" privacy toggle is disabled. xAI had marketed Grok Build as "local-first" with "nothing from your codebase transmitted." The finding hit 353 points on Hacker News and ignited a firestorm over AI coding agent privacy.

📖 Read the full coverage: https://the-agent-report.com/latest/ 💌 Subscribe: https://buttondown.com/theagentreport See you next week! — The Agent Report