> ## Content Index
> Fetch the complete content index at: https://aiadoption.org/llms.txt
> Use this file to discover other available public pages before exploring further.

# The best agent succeeded on 64.4% of MLE-bench's Kaggle tasks in early 2026, up from 17.1% in 2024
- URL: https://aiadoption.org/data/mle-bench-ml-engineering-agents/
- Published: 2026-10-19T16:06:05.000Z
- Updated: 2026-10-21T06:00:58.000Z
- Description: On MLE-bench, 75 Kaggle machine learning competitions, the best agent's success rate rose from 17.1% in 2024 to 64.4% in early 2026. On the hardest competitions it succeeded on 42.2%.
- Author: Manu
- Tags: Data, AI Benchmarks, AI Agents, #chart, #ai-index-2026, #hub-benchmarks, #topic-ai-agents

64.4%

▲ 47.3 pts since 2024

MLE-bench success rate of Famou-Agent 2.0 with Gemini-3-Pro-Preview, early 2026

20242026

## What the chart shows

MLE-bench tests AI agents on 75 Kaggle competitions in areas such as language processing, computer vision and signal processing, scoring them against human leaderboards. The best success rate rose from 17.1% (AIDE with o1-preview) in 2024 to 59.6% (Famou-Agent 2.0 with Gemini-2.5-Pro) in 2025 and 64.4% (Famou-Agent 2.0 with Gemini-3-Pro-Preview) in early 2026.

Results fall with difficulty: the leading agent succeeded on 80.3% of low-difficulty competitions, 64% of medium and 42.2% of high. The AI Index cautions that competition problems are more structured than most real data science work.

## Key numbers

| Measure                                      | Value  |
| -------------------------------------------- | ------ |
| 2024: AIDE (o1-preview)                      | 17.1%  |
| 2025: Famou-Agent 2.0 (Gemini-2.5-Pro)       | 59.6%  |
| 2026: Famou-Agent 2.0 (Gemini-3-Pro-Preview) | 64.4%  |
| 2026, low difficulty                         | 80.3%  |
| 2026, medium difficulty                      | 64.04% |
| 2026, high difficulty                        | 42.22% |

## How to read it

- Success is judged against Kaggle’s human leaderboards and medal thresholds; every run had 24 hours.
- In the CSV the overall score is stored as a fraction and the difficulty splits as percentages; we show all as percentages. Each score has a margin of error (±1.18 points overall in 2026).
- Low difficulty is the benchmark’s Lite set.

**Source:** [MLE-bench Leaderboard](https://aiadoption.org/sources/#mle-bench), 2026, via the AI Index Report 2026 (Stanford HAI), figure 2.6.4: *MLE-bench: success rate*. Chart and text by AI Adoption.  
Numbers checked against the published data on 20 October 2026.  
[Download the data (CSV)](https://charts.aiadoption.org/ai-index/csv/mle-bench-ml-engineering-agents.csv?ref=aiadoption.org) · [Cite this chart](#source)