Benchmark saturation in 2026: full ranking
Benchmarks where the top score is close to the ceiling leave less room to separate models. This table shows the top score, the index average and the gap between them.
What the data shows
Arena Elo has the highest top score at 1501.9, with a gap of 152.0 points over the average. A small gap means the benchmark is not separating the field much.
Saturation is a signal to look for harder or more task-specific evaluations. A benchmark that everyone passes stops being useful for model selection.
How to read it
- Prefer benchmarks with meaningful spread.
- Check whether the top score is near the ceiling.
- Pair saturated benchmarks with task-specific tests.
- Watch for contamination in widely used benchmarks.
Benchmark saturation
| Benchmark | Top score | Average | Gap | Category |
|---|---|---|---|---|
| Arena Elo | 1501.9 | 1349.9 | 152.0 | Preference |
| HumanEval | 96.8 | 74.0 | 22.8 | Coding |
| MATH-500 | 96.7 | 72.5 | 24.2 | Mathematics |
| HellaSwag | 96.1 | 81.4 | 14.7 | Commonsense |
| IFEval | 96.1 | 78.6 | 17.5 | Instruction following |
| MMLU-Pro | 91.6 | 73.5 | 18.1 | Knowledge |
| AIME 2025 | 91.4 | 63.6 | 27.8 | Mathematics |
| TruthfulQA | 91.1 | 73.6 | 17.5 | Safety |
| GPQA Diamond | 87.0 | 65.6 | 21.4 | Reasoning |
| MMMU | 86.9 | 67.1 | 19.8 | Multimodal |
| LiveCodeBench | 85.1 | 61.0 | 24.1 | Coding |
| SWE-bench Verified | 81.7 | 57.5 | 24.2 | Coding |
Methodology
This story uses the AI & LLM Index dataset: 600 models, 40 providers and 12 benchmarks. Prices are listed per million tokens. Scores are modelled estimates on a 0–100 scale and are not official benchmark results. Figures are rounded for readability.
Related
FAQ
What does the Benchmark saturation in 2026: full ranking data show?
Benchmarks where the top score is close to the ceiling leave less room to separate models. This table shows the top score, the index average and the gap between them.
Are the numbers official?
No. Prices and scores are compiled for this demo index. Benchmark values are modelled estimates, not official results. Verify figures with the provider or benchmark owner before making decisions.
How should I use this ranking?
Use it to form a shortlist. Check the specific model pages, compare price and context, and test the shortlist on your own examples before committing.