Chollet: Gemini 4 Neon's benchmark numbers 'really impressive at that price point'

fchollet · x · 2026-11-19

Keras creator and ex-Google star researcher François Chollet called the newly released Gemini 4 Neon "really impressive numbers at that price point," attaching Google DeepMind's official eval table: it leads Claude Opus 5.5 and GPT-6.2 Sol on most benchmarks — AA Index v5.0.1 of 61 (59/57), SWE-sweep provider harness 39.0% (10.1%/8.3%), AutomationBench 56.3%, GraphWalks long-context 95.2% (66.8%/80.1%), SEC-bench Pro 90.5%, WebExploitBench 51.3% (37.1%). It tops CursorBench 4.1 at 65.4%, though Claude Opus 5.5 edges it on Terminal-bench 4.1 (65.8% vs 65.2%) and FrontierCode 2.0 (47.3% vs 46.2%).

Related event: Google officially launches Gemini 4 Neon, its fastest frontier model yet(69 posts)→

Original post →

More from Models

Models channel →