Arena.ai (@arena)T1·Agents
58 pts
Arena.ai introduced its Coding Index, a benchmark measuring how well agents perform on real coding work. OpenAI's GPT-6.2 Astra (Max) scored 1338 points at #3 on the index, up 50 points from its predecessor while cutting median cost per task by 62% to $4.75, landing on the cost Pareto frontier. Its overall Agent Arena Index score was 1291 points at #6.