Models · Updated 16 Oct, 08:33 pm IST
Google DeepMind’s Gemini 4 Argon scores 97.5% on ARC-AGI-3 benchmark
Why it matters for readers: Gives a clear example of how researchers measure AI problem-solving and that scores can depend on how tests are run.
- Gemini 4 Argon scored 97.5% on the ARC-AGI-3 benchmark using the Provider Adapter Harness test setup.1
- The reported run cost about $4,800 and relied on vendor-specific features like retained reasoning and context compaction.1
- According to ARC Prize, DeepMind is the second major AI lab after OpenAI to cross 95% on ARC-AGI-3.1
- Argon matched GPT-6 Astra in needing 62% fewer actions than the human baseline across the 25 public ARC-AGI-3 games.1
- Epoch AI notes Argon trails rivals on tasks in math and software engineering, and results vary with testing method.1
Get a brief like this every morning
Uzha reads hundreds of sources and gives you the stories that matter for your work, with every source linked. Free.
Get started