According to Arthur Douillard, a member of Google's Gemini reinforcement learning team, within three months after SemiAnalysis published a critical report in August, Google released four Gemini 4 models—Argon, Carbon, Hydrogen, and Neon. Neon ranks third globally on Artificial Analysis' benchmark and leads in code generation evaluation, with average task completion costs about 50% lower than Claude Fable 5.2.
SemiAnalysis, which had predicted DeepMind was no longer a top-tier AI lab and could not catch up with leading competitors, later acknowledged the error. DeepMind founder Demis Hassabis responded to the apology, noting the analyst had underestimated the Google DeepMind team.