Arena.ai ranks GPT-6 Luna (Max) 23rd in Agent Arena

Arena.ai places GPT-6 Luna (Max) at No. 23 in Agent Arena, with a +1.6% net improvement across 8,000 real-world agentic sessions. The model is not on the Pareto frontier. Its median cost is $0.05 per task, 94% lower than GPT-6 Sol (Max) at $0.82 and 98% lower than GPT-6 Astra (Max) at $2.59.

The net-improvement score is within 0.09 percentage points of No. 22 GPT 5.5, while the median cost is 91% lower than that model’s $0.56 per task. Versus GPT-5.6 Luna (xHigh), at −0.9% and No. 29, this is a six-place point-rank move. Arena.ai’s clearest gains for GPT-6 Luna over GPT-5.6 Luna are Confirmed Success, No. 17 at +4.6% versus No. 33 at −4.6%, and Bash Recovery, No. 21 at +4.2% versus No. 27 at +2.2%.