Resolves to the lab whose model is ranked **#1 on the arena.ai Agent leaderboard** (https://arena.ai/leaderboard/agent, default view, sorted by Net Improvement) when I check it on **November 2, 2026, at or after 17:00 UTC**. I will save a Wayback Machine snapshot of the page at that time and link it in a comment. **Answers:** - **Anthropic**: any Claude model (Fable, Opus, Sonnet, Haiku, any effort setting) - **OpenAI**: any GPT / o-series model - **Google**: any Gemini or Gemma model (Google DeepMind) - **Any other lab**: everyone else (xAI/SpaceXAI, Meta, DeepSeek, Moonshot, Z.ai, Alibaba, Mistral, etc.) **State at creation (Oct 6, 2026):** #1 Claude Fable 5.1 (Max) at +14.31%, #2 Claude Opus 5.5 (High) +13.82%, #3 Claude Sonnet 5.5 (Max) +12.52%, #4 GPT 6 Astra (Max) +12.27%, #5 GPT 6.1 Sol (Max) +11.23%. Best Google: Gemini 4 Argon (High), #10 at +7.57%. 51 models listed. **Edge cases:** - Only the rank shown on the page counts. Preliminary/"new" badges don't matter; if the page lists a model at position 1, its lab wins. If two models are shown tied at #1, the one listed first on the page wins. - If the page is down on Nov 2, I use the nearest Wayback Machine snapshot taken between Nov 1 and Nov 3, 2026 (closest to 17:00 UTC Nov 2). - If the Agent leaderboard is discontinued or renamed, I use its direct successor on arena.ai; if there is none, the market resolves N/A. Arena notes that scores can move as more sessions come in and when new models join, so a new release in October could reshuffle the top even before it settles.