Google released Gemini 4 Argon on October 16, 2026, stating it is closer to the AI frontier than any previous Google model in months. It is the company's first major update to its Gemini series since the release of Gemini 3.1 Pro in February 2026.
Google reported a score of 97.5 percent on the ARC-AGI-3 benchmark, measured using the 'Provider Adapter Harness' setup, which allows vendor-specific features like retained reasoning and context compaction. That compares with the earlier 91.3 percent on ARC-AGI-2 and 98.5 percent on ARC-AGI-1.
Gemini 4 Argon is built on Google's advanced AI architecture and targets complex problem-solving tasks. Availability begins with a public release, initially for developers and researchers.
"Across the 25 public games of ARC-AGI-3, Gemini 4 Argon and GPT-6 Astra both needed 62 percent fewer actions than the human baseline," said Matthias Bastian, a researcher at the ARC Prize. The model often came close to the speedrun record in several games.
The announcement follows a July 2026 dispute over benchmark testing methods, where OpenAI's GPT-5.6 Sol outperformed Claude Opus 5 using advanced API settings. Google's latest model reflects improvements in testing conditions and performance.
Google did not say how Argon will perform in math and software engineering benchmarks, and it remains behind GPT-6 Astra and Opus 5.5 in some key areas. The company plans to continue refining the model for broader applications.
Source: thedecoder
