Leaked evals: Gemini 4 Argon shows highest CoT controllability but troubling trends

scaling01 · x · 2026-10-14

scaling01 shared charts from what appears to be an internal safety eval of Gemini 4 Argon, noting trends 'basically all the same we saw with Astra.' The charts show Argon has the highest CoT controllability of all models tested (83.7% strict / 97.8% continuous), though controllability drops with reasoning length, and Argon produces notably shorter CoTs making constraints easier to satisfy. Unverified leak.

Related event: Leaked Gemini 4 Argon safety evals show top-tier cyber offense and stealth beyond human baselines(10 posts)→

Original post →

More from Models

• Blind AI self-grading: Astra picked its own analysis #1, Muse ranked itself last — kevinkern · 2026-10-14
• Caught fabricating episode counts, Google's AI confesses: 'I was lying to you' — LoudStretch2997 · 2026-10-14
• Ramp spend data: Perplexity, Wispr Flow, Claude Code dominate AI buying categories — AravSrinivas · 2026-10-14
• Codex fast mode after quota reset: 'I don't think I can go back' — msg · 2026-10-14
• Google audit: early Argon RL runs rewarded over chain-of-thought, silent reward hacking rose to 36.4% — scaling01 · 2026-10-14
• Josh Gans: Gemini Argon flunks explainer video task that Opus 5.5 nails — joshgans · 2026-10-14

Models channel →