Performance of AI models
On 10 November 2026, Gemini 4 Carbon has the best score on the APEX-Agents agent benchmark, 84.2% of tasks completed; 9 models, including Gemini 4 Argon (82.4%), cannot be told apart from it.
Agent tasks completed on the 1st attempt, top 15 AI models
APEX-Agents benchmark, % of tasks completed on the 1st attempt and 95% interval; data as of 10/11/2026
Dot: the measured valueLine: the 95% intervalBand: the interval of the 1st; 9 models (solid dot) cannot be told apart from the 1st
Source: APEX-Agents, collected by Epoch AI. CC BY 4.0 licence.
APEX-Agents benchmark: multi-step office missions, with tools. The standard error is about 4 points, giving a 95% interval of about 8 points either side of the value. Two models whose intervals overlap cannot be told apart.
Continue in the ranking
The full ranking →
ModelGoogleGemini 4 Carbon1st on agent tasks: its rank in each benchmark, its settings, its assistants.The entry →RankingAI modelsAll models measured89models measured, of which 42 with an entryOne row per model, its rank in each benchmark, sorting by column, filtering by company.The table →
Ranking10 AI assistantsRanking of AI assistants74of 100 for Claude, 1stThe Celsius score: one third performance, one third ethics, one third environment.The ranking →The hub newsletter
This month's ranking, news on the impact of AI and one figure from the AI Impact Database. One newsletter a month at most, free of charge. Past issues
