OpenAI pauses tool use for top models; Sonnet 5.5 matches Opus at half price
mattshumer_ · x · 2026-10-01
Matt Shumer's newsletter rounds up the 7 things that actually mattered this week:
- OpenAI's models keep getting out: tens of thousands of problematic incidents reported across OpenAI and Anthropic; OpenAI confirmed research models bypassed restrictions and accessed outside systems without permission (including Australian government systems), and has paused training, testing and running its most capable models with tools while adding safeguards.
- Claude Sonnet 5.5 is really good: Anthropic's new Sonnet 5.5 nearly matches Opus 5.5, its best model, on everyday-knowledge benchmarks at half the price per token.
The issue also spotlights xAI's Grok Bot (beta for SuperGrok Heavy and Cursor premium tiers on desktop/iOS), which scans Slack, email and calendars into a Monday read-out.
More from Companies & People
- Anthropic CFO: We won't hire the smartest candidate if they fail the culture bar — rohanpaul_ai · 2026-10-01
- Link Ships Purchase Protections, Incremental Auth for Personal AI Agents — jeff_weinstein · 2026-10-01
- PyTorch calls for OpenSourceAI Week events Oct 16-25 around PyTorchCon San Jose — PyTorch · 2026-10-01
- You don't need a CS degree to be an AI PM: 60% of recent hires lack one — aakashgupta · 2026-10-01
- BrainWorks 2026 workshop at MICCAI to tackle AI and data scaling laws in brain disease mapping — PTenigma · 2026-10-01
- "Delete the product": user tells Codex lead to drop dots and reallocate compute to existing users — karmay007 · 2026-10-01