Arena.ai gates flagship LLMs behind battle mode

The decision to gate frontier models behind blind evaluations is a necessary pivot from a free playground to a sustainable research platform. High operational costs of "Superintelligence Tier" models like Claude Opus 4.6 made unrestricted direct access financially unviable for a research non-profit. Arena.ai is combatting "vibe coding" abuse where users utilize the site as a free API replacement for high-end models without contributing evaluation data. Forcing these models into "Battle Mode" ensures that every expensive token generated produces a human preference vote, reinforcing the leaderboard's statistical power. The platform's rebranding and this policy shift signal a maturation of the project as the industry's primary objective benchmark for LLM performance. Users seeking specific models by name must now shift to paid provider APIs, reflecting the end of the "unlimited free frontier" era for public benchmarking.