Theo: Gemini 4 Argon is a liability, keep it off your code
After a day of testing Gemini 4 Argon on his own codebases, Theo says it was factually wrong about more than half its claims in almost every test. That is a different kind of failure than a few points on a bench.
Receipts
Theo calls Gemini 4 Argon unusable after testing it on his codebases
Theo says Gemini 4 Argon aggressively hallucinates about how systems work and is unusable for real code work in his codebases. In almost every test he ran, he says the model was factually wrong about over 50% of its claims, and he walks through the methodology in a video.
This isn't the usual "oh it's 5% worse in this way, 3% better in others".
Theo lists 24 days of Opus output: 245 PRs, a Rust tsc, 1,200 commits
Asked by a skeptic what he had actually built, Theo itemized the 24 days since Opus dropped: a Rust rewrite of the TypeScript compiler he says is 2-3x faster than the official one, 245 PRs into T3 Code, 46 into Lakebed, and over 1,200 commits into lust-lang. He puts it at roughly 22 years of work in 21 days.
Is all of this real stuff that will ship to users? Absolutely not. But a lot of it is.
Theo: Codex still will not tell its own agent which model it is
Theo says Codex still does not put the current model in the system prompt, so the agent knows nothing about itself beyond "based on GPT-6". He says he has complained for a year straight and would trade a usage reset to get it fixed. He tagged an OpenAI account asking for it in the 28 days of Codex improvements.