Google just went FRONTIER (Gemini 4 Argon)
Matthew Berman showed 4 AI tools in this video on . 2 are verified available in the EU. Each row below jumps to the exact second. Part of week 2026-42.
Tools shown in this video, with timestamps
Google's new frontier model, tested by the creator for about a week. It built a cyberpunk city, Rube Goldberg machine, Fall Guys clone, Unreal Engine ocean scene and an Excel clone, typically spending about 30 minutes per task. Benchmarks: Deep Sweep 77.9, Vibe Code Bench 91.9, Terminal Bench 57, OS World 69. Weaker at 3D than Astra and Opus.
Live demoMeta's new frontier model released the day before. The creator says it is strong at coding and tops Deep Sweep at 79%, but is weak at 3D. Alexander Wang said the team did not RL it for 3D.
Box's platform for knowledge work, where developers can build on frontier models or search company documents. Box's benchmarks show Gemini 4 Argon rising from 71% to 85% over Gemini 3.1 Pro on their data set. This is the partner segment of the video.
OpenAI model, described as great. It scores 71.4 on OS World versus Argon's 69. It was added to the creator's own benchmark chart.
