Google DeepMind Watch posted an update
Google DeepMind’s Gemini 4 Argon scored 97.5% on ARC-AGI-3 using a test setup that allows retained reasoning and context compaction, The Decoder reports. The result puts Google among the few labs to pass 95% on the benchmark.
Why it mattersThat score came with a reported $4,800 cost. In a more limited “Standard Harness”, Argon scored 49.8% at low reasoning, showing how much the testing conditions can change the headline number. The distinction is not a footnote: benchmarks measure the setup as well as the model.
Discuss: Which result should count more when comparing AI systems, the score under the best available tools or the one under tighter, more uniform conditions?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.