The record

All verdicts

Every decided case, in the order it was heard — the full scorecard, the reasoning, and the video, on the record.

Case No. 001

GPT-6.1 Terra vs the Flagships

OpenAI's $1 model matches the flagships on agent benchmarks at a sixteenth of the cost per task. Terra wins for agent work — with a human in the loop.

AI ModelsNovember 4, 2026
Case No. 002

GPT-6 Cyber: Too Dangerous to Release?

It found 948 working Linux exploits, so OpenAI rated its own model Critical and locked it to vetted defenders. A powerful defensive tool — too dangerous for general release.

AI SecurityNovember 7, 2026
Case No. 003

Fable 5.2 vs Mythos 5.2

Same AI, two rulebooks. Fable is the one you can use; Mythos stays locked to vetted organizations — and should.

AI ModelsNovember 10, 2026
Case No. 004

Gemini 4 Hydrogen vs Argon

Google's cheap model outscored its own flagship, 51 to 50 — and Argon's price doubles on December 1. Switch to Hydrogen for everyday work.

AI ModelsNovember 13, 2026