Benchmarks · reasoning
DROP
Uses LLM Stats / ZeroEval DROP F1 leaderboard. Score: DROP F1. Collected by Jake Boggs' benchmark dashboard.
15 models · 15 results · data as of 2026-10-29 · llm-stats.com ↗
Leaderboard
DeepSeek V3 none
All results
15 rows| # | Model | Effort | DROP |
|---|---|---|---|
| 1 | DeepSeek V3 | none | 91.6% |
| 2 | Sonnet 3.5 | medium | 87.1% |
| 3 | MiMo V2.5 Pro | xhigh | 86.3% |
| 4 | GPT-4 | high | 86.0% |
| 5 | Llama 3.1 405b | none | 84.8% |
| 6 | GPT-4o | high | 83.4% |
| 7 | Haiku 3.5 | medium | 83.1% |
| 8 | Opus 3 | none | 83.1% |
| 9 | GPT-4o mini | high | 79.7% |
| 10 | Llama 3.1 70b | none | 79.6% |
| 11 | Sonnet 3 | none | 78.9% |
| 12 | Haiku 3 | medium | 78.4% |
| 13 | Gemini 1.5 Pro | none | 74.9% |
| 14 | GPT-3.5 | high | 70.2% |
| 15 | Llama 3.1 8b | medium | 59.5% |