Bob M

Key Takeaways - V4 Pro output tokens: $0.87/M, permanent as of May 23. That's 34x cheaper than GPT-5.5's $30/M and 29x cheaper than Claude Opus 4.7's $25/M. - Input tokens run $0.435/M. Cache hits? Just $0.

Only 97 out of more than 10,000 critical vulnerabilities discovered by Anthropic's unreleased Claude Mythos Preview model have been patched. The rest are sitting in disclosed-but-unfixed limbo, sitting in code you probably run in production right now. Let that number land. Mythos Preview, distributed to roughly 50

Key Takeaways - Locked Use lets Codex work on a dark, locked Mac. No human at the keyboard needed - Accessibility + Screen Recording permissions give it near-total UI control; same permissions a RAT would want - EU, UK, Switzerland blocked at launch. That tells you something - SOC 2 clients?

Key Takeaways - Command A+ hits 218B params but only fires 25B per token. Hence two H100s, 4-bit quantized, first Apache 2.0 model from a Western AI lab. - Benchmarks jump: 85% on Telecom agent tasks (vs. 37% before), 25% harder coding wins, sits at 37 on the Artificial

- GPT-4.5 was judged human 73% of the time in a UC San Diego PNAS study. Beating the actual human in the room. - Without a persona prompt, the same model scored 36%. One paragraph of instructions doubled the pass rate. - GPT-5 still passed at 59% even with

Key Takeaways - May 20, 2026: a general-purpose OpenAI model published a complete, verified disproof of the planar unit distance conjecture. An open problem Paul Erdős handed to mathematics in 1946. - Timothy Gowers, Fields Medalist, endorsed the proof without caveats. Princeton's Will Sawin called it the most