Equation-Suffix Prediction (Kimi K2.6 Scorer): leaderboard
Forecasting the rest of a displayed equation in recent arXiv physics and mathematics papers (Ranard, 2026): the model reads 10,000 characters of context and the start of the equation and writes a forecast, and the fixed Kimi K2.6 scorer measures how much that forecast raises the clipped per-token log-likelihood (clipLL2) of the true suffix over a same-budget recent-context control. Mean lift over 1,363 equation cuts from 138 papers, on a scale from -2 to 2; higher is better.
Metric: Likelihood lift (mean clipLL2 per target token over the same-budget recent-context control, range -2 to 2; 1,363 equation-suffix cuts from 138 recent arXiv physics and mathematics papers; Kimi K2.6 scorer). Source: arxiv.org. Saturation forecast: Around June 2028. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 (High) | 0.17 | #48 (GPT-5.5) |
| 2 | GPT-5.5 (Medium) | 0.17 | #48 (GPT-5.5) |
| 3 | GPT-5.5 (Low) | 0.16 | #48 (GPT-5.5) |
| 4 | Claude Opus 4.7 (Medium) | 0.15 | #75 (Claude Opus 4.7) |
| 5 | Claude Opus 4.7 (Low) | 0.15 | #75 (Claude Opus 4.7) |
| 6 | GPT-5.5 (Non-reasoning) | 0.13 | #48 (GPT-5.5) |
| 7 | GPT-5.4 Nano (High) | 0.08 | #633 (GPT-5.4 Nano) |
| 8 | GPT-5.4 Nano (Medium) | 0.07 | #633 (GPT-5.4 Nano) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=equation-suffix-prediction-kimi-k2-6-scorer · How It Works · Data refreshed daily, snapshot 2026-11-12.