Understand this story
ARC Prize released validation results for Gemini 4 Hydrogen: the model scored 92.5% on ARC-AGI-2 and 98.5% on ARC-AGI-1, respectively, matching or outperforming Gemini 3.8 Flash at the high reasoning level while using approximately 52% fewer reasoning tokens; the per-task costs for the two tests were $1.26 and $0.18, respectively. ARC Prize also said that ARC-AGI-3 testing could not be completed due to API issues involving multi-action function calling, MALFORMED_FUNCTION_CALL, and other problems. There are currently no other evidence-backed subsequent developments.
AI-generated from public reports
Latest development · Report
ARC Prize Releases ARC-AGI Verification Results for Gemini 4 HydrogenDevelopments
1 confirmed developments
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Nov 21
ARC PrizeFirst-party
ARC Prize Releases ARC-AGI Verification Results for Gemini 4 Hydrogen
Gemini 4 Hydrogen scored 92.5% and 98.5% on ARC-AGI-2 and ARC-AGI-1, respectively, at a per-task cost of $1.26 and $0.18. At the high reasoning level, it matched or outperformed Gemini 3.8 Flash while using approximately 52% fewer reasoning tokens. ARC Prize was unable to complete ARC-AGI-3 testing due to API issues involving multi-action function calls, MALFORMED_FUNCTION_CALL errors, and related problems.
AI Trends