OpenAI·released 2023-03-141 source
GPT-4 (Mar 2023) benchmark scores: 7 benchmarks tracked, leading 2. 29% of its scores are independently reproduced — the rest are self-reported or unverified.
Leads on 2 of 7 benchmarks·best result 93.73 on HellaSwag (unverified)·2 of 7 independently reproduced
Head-to-headGPT-4 (Mar 2023) vs GPT-4o mini| Benchmark | Score | Evidence status | Measured |
|---|---|---|---|
| HellaSwag | 93.73accuracy (%) | unverified· optimizedT1 | 2023-03-14The Falcon Series of Open Language Models |
| GSM8K | 92accuracy (%) | self-reported· optimizedT1 | 2023-03-14GPT-4 Technical Report |
| MMLU | 81.87accuracy (%) | self-reported· optimizedT1 | 2023-03-14GPT-4 Technical Report |
| Winogrande | 75accuracy (%) | self-reported· optimizedT1 | 2023-03-14GPT-4 technical report |
| METR Time Horizons | 36.11accuracy (%) | unverifiedT1 | 2023-03-14METR - Measuring AI Ability to Complete Long Tasks |
| GPQA diamond | 14.31accuracy (%) | reproduced· optimizedT1 | 2023-03-14Epoch AI |
| OTIS Mock AIME 2024-2025 | 0.46accuracy (%) | reproduced· optimizedT1 | 2023-03-14Epoch AI |
Claims drawn from cited facts, not live model generation.
This model originates from OpenAI and was released in 2023.
2 cited facts
This model is evaluated on seven benchmarks. It currently holds the top score on two of those. Because the scores have been independently reproduced, the record is credible.
3 cited facts
It trails the leader by 32.39 points on average, a wide gap that places it well behind the front of the pack. It does hold the top score on 2 benchmarks, but that category leadership is isolated against an otherwise large average deficit.
2 cited facts
1 facts cross-checked across data sources: 1 single-source.
Source facts, citations, and refresh stamp for this record.
Sources: epoch_benchmarks
GPT-4 (Mar 2023) is an AI model developed by OpenAI, released 2023-03-14. Its benchmark record below tags every score as independently reproduced, self-reported, or unverified.
GPT-4 (Mar 2023) has recorded scores on 7 benchmarks, each shown with its evidence status.
GPT-4 (Mar 2023) has recorded scores on 7 benchmarks — Winogrande, GPQA diamond, OTIS Mock AIME 2024-2025, MMLU, METR Time Horizons, GSM8K, and 1 more. The full table above shows each score with its evidence status.
2 of 7 recorded scores (29%) are independently reproduced rather than self-reported by the lab.
GPT-4 (Mar 2023) has 7 tracked claims: 2 independently reproduced, 3 self-reported, 2 unverified.