The open versus proprietary gap in 2026: full ranking
Average scores for open-weight and proprietary models on each benchmark in the index.
What the data shows
The gap varies by benchmark. On some tasks the two groups are close; on others the proprietary average is clearly ahead. A single aggregate number hides that variation.
The comparison is descriptive, not causal. Group averages mix different model sizes, prices and release dates.
How to read it
- Compare the gap on the benchmark closest to your task.
- Check the best open model, not just the group average.
- Account for self-hosting cost.
- Test both groups on your own data.
Open versus proprietary scores
| Benchmark | Open avg | Proprietary avg | Gap |
|---|---|---|---|
| MMLU-Pro | 69.7 | 79.7 | 10.0 |
| GPQA Diamond | 61.1 | 72.9 | 11.9 |
| SWE-bench Verified | 52.3 | 66.0 | 13.6 |
| HumanEval | 69.2 | 81.8 | 12.6 |
| MATH-500 | 67.2 | 81.0 | 13.8 |
| AIME 2025 | 57.5 | 73.5 | 16.0 |
| LiveCodeBench | 55.4 | 70.2 | 14.8 |
| IFEval | 74.7 | 84.9 | 10.1 |
| TruthfulQA | 69.7 | 79.9 | 10.1 |
| HellaSwag | 78.3 | 86.4 | 8.1 |
| MMMU | 63.0 | 73.9 | 11.0 |
| Arena Elo | 1315.2 | 1406.2 | 91.0 |
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 The open versus proprietary gap in 2026: full ranking data show?
Average scores for open-weight and proprietary models on each benchmark in the index.
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.