PlatinumBench (MIT): leaderboard
PlatinumBench live text leaderboard: mean error rate with equal weight across five categories of 14 curated datasets. Lower is better. Scores use category means of per-dataset error percentages, not pooled error counts; source checkpoint identifiers and reasoning settings are retained.
Metric: Category-balanced mean error rate (%). Source: huggingface.co. Source: platinum-bench.csail.mit.edu. Saturation forecast: Estimated already saturated. 34 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro (Preview 03-25) | 0.53 | #123 |
| 2 | O1 (2024-12-17) (High) | 0.74 | #149 (O1 (2024-12-17)) |
| 3 | Grok 3 | 0.79 | #304 |
| 4 | GPT-4.5 (Preview) | 0.99 | #244 |
| 5 | Claude 3.7 Sonnet (20250219) | 1.05 | #201 |
| 6 | Claude 3.5 Sonnet (20241022) | 1.09 | #291 |
| 7 | GPT-4.1 (2025-04-14) | 1.11 | #250 |
| 8 | O1 (Medium) | 1.27 | #183 (O1) |
| 9 | DeepSeek R1 | 1.4 | #251 |
| 10 | Grok 3 Mini (High) | 1.51 | #281 (Grok 3 Mini) |
| 11 | Claude 3.5 Sonnet (20240620) | 1.84 | #299 |
| 12 | Llama 3.1 405B Instruct | 1.92 | #457 |
| 13 | Qwen 2.5 Max | 2.05 | #416 |
| 14 | Llama 4 Maverick Instruct FP8 | 2.12 | #414 |
| 15 | GPT-4o (2024-08-06) | 2.24 | #336 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=platinumbench-mit · How It Works · Data refreshed daily, snapshot 2026-10-20.