About LLM Verifications Leaderboard
LLM Verifications Leaderboard is an independent research project by Suhas Bhairavthat benchmarks large language models, published as part of his portfolio at suhasbhairav.com. It tests every model through OpenRouter's public API with the same expert-level multiple-choice questions from open datasets and the same streamed speed test, and publishes accuracy, output speed, time to first token and billed cost per question. Every request and response is recorded and hashed, the statistical uncertainty of each score is stated, and no lab or host pays for placement.
Key facts
| Models tested | 78 |
|---|---|
| AI labs | 28 |
| API hosts | 70 |
| Questions per model | 80 |
| Source datasets | SuperGPQA, MMLU-Pro |
| Latest run | 27 Oct 2026 |
| Founded | 2026 |
| Run by | Suhas Bhairav |
How to cite the results
Please link to the page the number comes from and give the test date shown on it, since results change as models are re-tested. Suggested format:
LLM Verifications Leaderboard (2026). AI model benchmark results. https://suhasbhairav.com/research/llm-leaderboard/leaderboard (data as of 27 Oct 2026).
For a single model or comparison, replace the address with that page, for example a model page under https://suhasbhairav.com/research/llm-leaderboard/models/ or a comparison under https://suhasbhairav.com/research/llm-leaderboard/compare/.
Describing the numbers accurately
- The verification score is the share of 80 multiple-choice questions answered correctly, not a general intelligence rating.
- Scores a few points apart are statistically tied; the methodology gives the uncertainty.
- Cost per question is what the API billed, reasoning tokens included, on the test date.
Press contact
For interviews, data requests or questions about a result, email suhasbhairav@gmail.com. Please mention your deadline. Background on how the leaderboard works is on the about page and in the editorial standards.