GPT-4 Jun 2023
GPT-4 ranks first on 3 of the 39 benchmarks with five or more models scored.
44
Benchmarks with a score
3
Ranked first, out of five or more models
5
In the top three
OpenAI Jun 2023
GPT-4
Jun 2023 snapshot Closed weights
120.1
#184 of 256
Epoch Capabilities Index
- GPQA Diamond 30.6% #207 /229
- BigCodeBench-Hard 17.6% #65 /96
- ToolComp 51.4% #23 /32
Where it ranks
Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.
Knowledge and language
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| EQ-Bench v2 | 84.8 | #6 of 85 | Official leaderboard | |
| Vectara Hallucination Leaderboard (HHEM) | 1.80% | #17 of 148 | Official leaderboard | |
| GAOKAO-Bench | 71.6% | #1 of 7 | Official leaderboard | |
| TriviaQA | 84.8% | #7 of 38 | Collected by Epoch AI, source not named | |
| BIG-Bench Hard | 75.1% | #9 of 45 | Claim in Baichuan 2: Open Large-scale Language Models, via Epoch AI | |
| AlpacaEval | 93.8% | #6 of 19 | Official leaderboard | |
| Chinese SimpleQA | 47.4% | #9 of 27 | Official leaderboard | |
| ChatRAG Bench | 53.9 | #1 of 3 | Official leaderboard | |
| AlpacaEval 2.0 | 30.2% | #16 of 47 | Official leaderboard | |
| Arena-Hard-Auto | 37.9% | #21 of 38 | Official leaderboard | |
| LMArena (Elo) | 1,276 | #249 of 335 | Official leaderboard | |
| Multilingual human evaluation | 3.71 | #5 of 5 | Official leaderboard |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| GSM8K | 90% | #7 of 70 | Claim in Baichuan 2: Open Large-scale Language Models, via Epoch AI | |
| DTBench | 62.7% | #132 of 180 | Official leaderboard | |
| ForecastBench | 0.15 | #63 of 82 | Official leaderboard | |
| MATH Level 5 | 23% | #73 of 94 | Epoch AI run | |
| GPQA Diamond | 30.6% | #207 of 229 | Epoch AI run | |
| OTIS Mock AIME 2024-2025 | 1.11% | #202 of 212 | Epoch AI run |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| CodeEditorBench | 85.5% | #1 of 7 | Official leaderboard | |
| BigCodeBench | 51.6% | #15 of 69 | Official leaderboard | |
| CRUXEval | 77.1% | #3 of 11 | Official leaderboard | |
| Aider-Edit | 67.7% | #16 of 34 | Official leaderboard | |
| DS-1000 | 51% | #2 of 4 | Official leaderboard | |
| BigCodeBench-Hard | 17.6% | #65 of 96 | Official leaderboard | |
| Bird-SQL | 54.9% | #9 of 13 | Official leaderboard | |
| SWE-bench Lite | 23.7% | #11 of 15 | Official leaderboard | |
| WeirdML | 12.4% | #108 of 112 | Official leaderboard | |
| TasteVal | 2.72× | #20 of 20 | Official leaderboard |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| AgentBench | 4.01 | #1 of 5 | Official leaderboard | |
| GAIA | 64.5% | #25 of 64 | Official leaderboard | |
| WebArena | 14.9% | #2 of 4 | Official leaderboard | |
| ToolComp | 51.4% | #23 of 32 | Official leaderboard | |
| Mystery Game Puzzles | 12% | #66 of 84 | Epoch AI run |
Professional work
Vision, audio and robotics
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| OpenEQA | 33.5 | #2 of 2 | Official leaderboard |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SuperCLUE-Safety | 87.4 | #2 of 7 | Official leaderboard | |
| nocot-bench (NCRI) | 72.8 | #84 of 185 | Official leaderboard | |
| SORRY-Bench | 28.9% | #12 of 23 | Official leaderboard | |
| Situational Awareness Dataset (SAD) | 46.7% | #9 of 17 | Official leaderboard | |
| AIR-Bench 2024 | 64.2% | #45 of 82 | Official leaderboard |
Other
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| HELM Lite | 86.7 | #6 of 71 | Official leaderboard | |
| Conceptual Reasoning Index | 25.3 | #125 of 151 | Official leaderboard | |
| ACCoRD | 37.3 | #136 of 152 | Official leaderboard |
OpenAI's models over time
OpenAI's models in color, against every other model on the Epoch Capabilities Index, by release date.
Source: Epoch Capabilities Index, Epoch AI (CC BY 4.0)