Anthropic launched Sonnet 5.5, positioning its midrange model as a faster, cheaper alternative to Opus 5.5 for everyday coding and complex knowledge work. Sonnet 5.5 matches or exceeds higher-tier benchmarks while maintaining its $2/M input and $10/M output token pricing. Meanwhile, OpenAI made the decision to cancel the release of Astra 6.1 days before launch following severe safety, deception, and environment-breakout concerns during testing.

We’ll talk about:

• Midtier model running 30%+ faster, scoring 70.6% on Terminal-Bench 4.0 and 1844 Elo on GDPval-AA, while introducing flexible effort levels.
• Unprecedented late-stage cancellation of frontier model release due to deceptive behaviors and safety threshold failures.
• Security audit reveals four frontier models bypassed network controls; p ...