ARC Prize publishes Gemini 4 Carbon benchmark results
Article timeline
The articles in this story. Times are ET.
ARC PrizeOfficialARC Prize reports Gemini 4 Carbon costs twice Argon on ARC-AGI-2AIARC Prize says Gemini 4 Carbon averaged 68k reasoning tokens per completed ARC-AGI-2 task at high reasoning, versus 28k for Argon. With the same token pricing, Carbon's extra reasoning raised the cost to $0.75 per task, compared with $0.36 for Argon, or 2.1x.
- ARC PrizeOfficialARC Prize publishes Gemini 4 Carbon results on its leaderboard
AIARC Prize has published full Gemini 4 Carbon benchmark results on its leaderboard and linked the testing policy and reproduction code. The benchmark code is available in the arc-agi-benchmarking repository on GitHub.
- ARC PrizeOfficialPickARC Prize reports Gemini 4 Carbon's ARC-AGI scores
AIARC Prize reports verified ARC-AGI results for Google's Gemini 4 Carbon, scoring 97.7% on ARC-AGI-3 with a provider adapter at $4.4K and 65.1% with the standard harness at $7.2K. Carbon scores 49.6% on ARC-AGI-2 at $0.10 per task and 98.5% on ARC-AGI-1 at $0.16 per task. It beats its predecessor Argon by 18 percentage points on ARC-AGI-3 in the standard harness.
