InfiniteBench: leaderboard
InfiniteBench primary arXiv 2402.13718v3 Table 3 published Average across 12 tasks. GPT-4 Turbo API run and Kimi web run retain source-specific identities because exact snapshots are unspecified; Claude 2 web and the explicit Yarn-Mistral-7B-128k checkpoint remain distinct. Replaces unsupported midpoint averages from the censored README.
Metric: Published Average Score (%). Source: github.com. Source: arxiv.org. Saturation forecast: Around January 2027. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Yarn-Mistral-7B-128k | 19.96 | #1398 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=infinitebench · How It Works · Data refreshed daily, snapshot 2026-10-20.