Arena announces GPT-6 Astra testing in Battle Mode and Agent Mode

Arena announced on September 4 that OpenAI’s GPT-6 Astra was available for testing in Battle Mode and Agent Mode. Arena says its Agent Arena evaluates models on millions of real-world, long-horizon tasks, where models can access web search, filesystem and terminal tools. It describes the leaderboard as measuring outcomes relative to the average model using a causal tracing methodology.