A post calls GDM’s results impressive and notes a “backwards bend” in action efficiency.
A post says Argon is just the second model family to officially, effectively max out ARC-AGI-3 using a provider adapter. It praises GDM’s results and says the same “backwards bend” is appearing in action efficiency.
100K
16 Sources, first seen ago
Gemini 4 Argon scored 97.5% on ARC-AGI-3 using a provider-adapter harness, while its standard-harness score was 49.8%, according to verified results posted by ARC Prize on Oct. 15, 2026. The two figures describe different evaluation setups, not interchangeable scores for the same configuration.
ARC Prize described Google DeepMind as the second frontier AI lab to score above 95% on ARC-AGI-3. The result above that threshold is the provider-adapter score; it is not a perfect 100% result or the standard-harness result.
The same post lists Argon at 91.3% on ARC-AGI-2, with a cost of $0.32 per task, and 98.5% on ARC-AGI-1, at $0.16 per task.
In a quote-post of the results, Mike Knoop characterized Argon as the second model family to effectively max out ARC-AGI-3 using a provider adapter. He praised Google DeepMind's results and highlighted what he called a “backwards bend” in action efficiency.
Google DeepMind's team says it built the 12-degree-of-freedom robot from one plain-English prompt without manual CAD.
In an October 28 post, Li recalled work on AlphaCode and Gemini and outlined plans for a break.
Google DeepMind's team says it built the 12-degree-of-freedom robot from one plain-English prompt without manual CAD.
In an October 28 post, Li recalled work on AlphaCode and Gemini and outlined plans for a break.
@arcprizeGemini 4 Argon from @Google on ARC-AGI (Verified): - ARC-AGI-3: 97.5%, $4.8K (provider adapter harness), 49.8%, $5.8K (standard harness) - ARC-AGI-2: 91.3%, $0.32/task - ARC-AGI-1: 98.5%, $0.16/task Google DeepMind is the second frontier AI lab to score above 95% on ARC-AGI-3.
@mikeknoopArgon is just the second model family to officially, effectively, max out ARC-AGI-3 using a provider adapter. Impressive results from GDM here! Also notable that we're seeing the same "backwards bend" on action efficiency.
@scaling01I'm not quite sure what role loopers played in saturating ARC-AGI-3 but I definitely believe it's solvable by just a shit ton of RL on games. loopers probably just made it a lot cheaper overall I think I am at 85% RL on games and 15% looping for what caused this sudden jump, mostly because of the Opus 5 datapoint and general trends in AI (everything can be RL'd)@arcprizeGemini 4 Argon from @Google on ARC-AGI (Verified): - ARC-AGI-3: 97.5%, $4.8K (provider adapter harness), 49.8%, $5.8K (standard harness) - ARC-AGI-2: 91.3%, $0.32/task - ARC-AGI-1: 98.5%, $0.16/task Google DeepMind is the second frontier AI lab to score above 95% on ARC-AGI-3.@mikeknoopArgon is just the second model family to officially, effectively, max out ARC-AGI-3 using a provider adapter. Impressive results from GDM here! Also notable that we're seeing the same "backwards bend" on action efficiency.@scaling01I'm not quite sure what role loopers played in saturating ARC-AGI-3 but I definitely believe it's solvable by just a shit ton of RL on games. loopers probably just made it a lot cheaper overall I think I am at 85% RL on games and 15% looping for what caused this sudden jump, mostly because of the Opus 5 datapoint and general trends in AI (everything can be RL'd)@arcprizeGemini 4 Argon from @Google on ARC-AGI (Verified): - ARC-AGI-3: 97.5%, $4.8K (provider adapter harness), 49.8%, $5.8K (standard harness) - ARC-AGI-2: 91.3%, $0.32/task - ARC-AGI-1: 98.5%, $0.16/task Google DeepMind is the second frontier AI lab to score above 95% on ARC-AGI-3.@mikeknoopArgon is just the second model family to officially, effectively, max out ARC-AGI-3 using a provider adapter. Impressive results from GDM here! Also notable that we're seeing the same "backwards bend" on action efficiency.@scaling01I'm not quite sure what role loopers played in saturating ARC-AGI-3 but I definitely believe it's solvable by just a shit ton of RL on games. loopers probably just made it a lot cheaper overall I think I am at 85% RL on games and 15% looping for what caused this sudden jump, mostly because of the Opus 5 datapoint and general trends in AI (everything can be RL'd)@arcprizeGemini 4 Argon from @Google on ARC-AGI (Verified): - ARC-AGI-3: 97.5%, $4.8K (provider adapter harness), 49.8%, $5.8K (standard harness) - ARC-AGI-2: 91.3%, $0.32/task - ARC-AGI-1: 98.5%, $0.16/task Google DeepMind is the second frontier AI lab to score above 95% on ARC-AGI-3.@mikeknoopArgon is just the second model family to officially, effectively, max out ARC-AGI-3 using a provider adapter. Impressive results from GDM here! Also notable that we're seeing the same "backwards bend" on action efficiency.@scaling01I'm not quite sure what role loopers played in saturating ARC-AGI-3 but I definitely believe it's solvable by just a shit ton of RL on games. loopers probably just made it a lot cheaper overall I think I am at 85% RL on games and 15% looping for what caused this sudden jump, mostly because of the Opus 5 datapoint and general trends in AI (everything can be RL'd)—
Not ranked yet
—
Not ranked yet
