Agent Arena Benchmark Updates: GPT-5.6 Sol, Terra, and Luna Show Performance Gains

Agent Arena has reported performance improvements for several AI models, with GPT-5.6 Sol securing the second spot on the leaderboard with a 10.1% increase. Additionally, the Terra (+4.0%) and Luna (+3.3%) models achieved positive net improvements, placing them close to Claude Opus 4.8 (+3.5%). These updates highlight the rapid evolution of AI agents in handling complex, dynamic tasks rather than static benchmarks. The close competition between GPT-5.6 variants and Claude Opus 4.8 demonstrates the narrowing gap among top-tier agent models. The evaluations were conducted at the "xHigh" setting within Agent Arena, a platform designed to test agents in live, interactive environments. The benchmark prevents models from relying on memorization by evaluating them on real-world workflows like coding and browsing.