TPS-CalcBench - Exact Answers: leaderboard
Metric: Share of items (%) whose predicted value is within 1 percent relative error of the reference value (weighted over target quantities; missing answers count as wrong) on the 420-item high-confidence core set of TPS-CalcBench (analytical calculation problems in hypersonic thermal protection system engineering), zero-shot text-only prompting; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 12 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.2 | 51.9 |
| 2 | GPT-5.1 | 48.1 |
| 3 | GLM-5 | 47.1 |
| 4 | DeepSeek V3.1 | 43.1 |
| 5 | Claude Opus 4.5 | 41.9 |
| 6 | Seed 1.8 | 38.6 |
| 7 | GLM-4.7 | 37.6 |
| 8 | Claude Sonnet 4.5 | 33.6 |
| 9 | Claude Haiku 4.5 | 24.3 |
| 10 | Gemini 3 Flash (Preview) | 9.8 |
| 11 | MiniMax-M2.5 | 0 |
| 12 | MiniMax-M2.1 | 0 |
Interactive version: theaggregate.ai/benchmark?slug=tps-calcbench-exact-answers · How It Works · Data refreshed daily, snapshot 2026-10-07.