Gemini 4 Carbon takes 3x longer to complete

micjm.com1 min read

An analysis of benchmark data reveals that the Gemini 4 Carbon AI model is significantly slower than its opponents, taking roughly three times as long to complete tasks. The author notes that Carbon's poor performance is driven by higher output token counts and increased tool calls rather than inference speed or latency. Despite its slow completion times, Carbon maintains a high win rate in final evaluations when voters wait for both solutions to finish. Open article →

More on this topic