Benchmark
ACEBench
Agentic benchmarks measure how well models carry out tasks that take several steps. Scores are percentages.
All agentic benchmarksScores from ACEBench data, LLM Stats data.
- Models measured
- 18
- Scores recorded
- 18
- Sources
- 2
- IQ weight
- 0.85
Score distribution
- 62.6%
- Mean
- 70.7%
- Median
- 18.8%
- Std dev
- 33.4%
- 10th percentile
- 80.6%
- 90th percentile
- 19.6%
- Min
This benchmark has a weight of 0.85 for chat models, measured over 18 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.
Leaderboard
| # | Model | Score |
|---|---|---|
| 1 | GPT 4o· ACEBench data | 85.4% |
| 2 | Kimi K2· ACEBench data | 82.2% |
| 3 | Gemini 2.5 Pro· ACEBench data | 80.6% |
| 4 | Qwen 2.5 Coder 32B Instruct· ACEBench data | 79.6% |
| 5 | Qwen Max· ACEBench data | 78.4% |
| 6 | Kimi K2 Instruct· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 76.5% |
| 7 | Qwen2.5 72B Instruct· ACEBench data | 74.7% |
| 8 | GPT-4o mini· ACEBench data | 72.5% |
| 9 | Gemini 1.5 Pro· ACEBench data | 70.7% |
| 10 | Claude 3.5 Sonnet· ACEBench data | 68.9% |
| 11 | Llama 3.1 70B Instruct· ACEBench data | 60.4% |
| 12 | Doubao Pro 32K· ACEBench data | 59.4% |
| 13 | Qwen2.5 7B Instruct· ACEBench data | 54.8% |
| 14 | deepseek-coder-v2-lite-instruct· ACEBench data | 49.5% |
| 15 | Qwen2.5 Coder 7B Instruct· ACEBench data | 48.9% |
| 16 | Llama 3.1 8B Instruct· ACEBench data | 33.4% |
| 17 | Phi-3 Mini 128K Instruct· ACEBench data | 32% |
| 18 | Llama 3.2 3B· ACEBench data | 19.6% |
Data credits: ACEBench, MIT (licence) · Data by LLM Stats