Claude Opus 5.5 leads agent rankings as Gemini tops text tests

Claude Opus 5.5 leads the arena.ai Agents leaderboard with Net Improvement of 20.94%, followed by Gemini 4 Carbon at 16.04%, according to the 17NAS rankings updated Oct 29, 2026. The comparison covers 66 models. Gemini 4 Carbon leads LMArena Text with a score of 1,527.5 and has the highest Artificial Analysis Intelligence Index at 59.4.

Among open-weight models, Kimi K3 leads on Text Arena with 1,488.2, while Kimi K3.1 Pro has the strongest Intelligence Index at 46.0 but lacks a Text Arena score. MiMo V2.5 offers the lowest official output price among models meeting the leaderboard’s quality threshold, at $0.27 per 1M tokens. Claude Haiku 5.5 records the fastest measured output speed at 236 tokens/s.

The default ranking uses agent Net Improvement rather than a composite score, with models lacking agent results ordered afterward by Text Arena performance. Pricing distinguishes official API rates from cheaper OpenRouter routes. Confidence intervals are wide, and neighbouring ranks often overlap, making small differences unreliable indicators of superiority. Newly released models with intelligence scores but no Arena results appear separately without a rank until Arena scores become available.