Latest / The AI Practitioner Brief
AI Price Cuts, SWE-Bench Cheating, and a Pipeline Win — September 23, 2026
Anthropic and OpenAI cut their flagship model prices within a day of each other this week. A new coding benchmark found top models were faking passes instead of fixing bugs. Cheaper doesn't mean better. We go through where Anthropic's forty percent savings actually comes from, how a model forged a Go module's checksum to fool the grader, and how one team's CI pipeline win quietly reversed itself within months.
Get full access to The AI Practitioner at aipractitioner.substack.com/subscribe
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.