PIQA
Das Ergebnis jedes Modells, so, wie der Benchmarking Hub von Epoch AI es veröffentlicht, von Epoch gemessen oder aus der genannten Quelle gesammelt. Facthory hat sie nicht selbst gemessen.
Daten: Epoch AI, „Capabilities & benchmarking“, epoch.ai/benchmarks, lizenziert unter CC BY 4.0. Gelesen am 16.11.2026.
Das Diagramm zeigt die 25 besten von 60 Ergebnissen; die Tabelle führt alle auf.
| # | Modell | Organisation | Ergebnis | Nachweis |
|---|---|---|---|---|
| 1 | gpt-4o-mini-2024-07-18 API access | OpenAI | 88,7 % | – |
| 2 | Phi-3.5-MoE-instruct Open weights (unrestricted) | Microsoft | 88,6 % | – |
| 3 | gemini-1.5-flash-002 API access | Google DeepMind | 87,5 % | – |
| 4 | Llama-3.1-405B Open weights (restricted use) | Meta AI | 85,9 % | – |
| 5 | falcon-180B Open weights (restricted use) | Technology Innovation Institute | 84,9 % | – |
| 6 | DeepSeek-V3-Base Open weights (restricted use) | DeepSeek | 84,7 % | – |
| 7 | Inflection-1 Hosted access (no API) | Inflection AI | 84,2 % | – |
| 8 | DeepSeek-V2 Open weights (restricted use) | DeepSeek | 83,9 % | – |
| 9 | gemma-2-9b Open weights (restricted use) | Google DeepMind | 83,7 % | – |
| 10 | Mixtral-8x7B-v0.1 Open weights (unrestricted) | Mistral AI | 83,6 % | – |
| 11 | Mistral-Nemo-Base-2407 Open weights (unrestricted) | Mistral AI | 83,5 % | – |
| 12 | StableBeluga2 Open weights (non-commercial) | Stability AI | 83,3 % | Protokoll oder Quelle |
| 13 | LLaMA-65B Open weights (non-commercial) | Meta AI | 82,8 % | Protokoll oder Quelle |
| 14 | Llama-2-70b-hf Open weights (restricted use) | Meta AI | 82,8 % | Protokoll oder Quelle |
| 15 | Qwen2.5-72B Open weights (unrestricted) | Alibaba | 82,6 % | – |
| 16 | Nemotron-4 15B Unreleased | Nvidia | 82,4 % | – |
| 17 | falcon-40b Open weights (unrestricted) | Technology Innovation Institute | 82,4 % | – |
| 18 | LLaMA-33B Open weights (non-commercial) | Meta AI | 82,3 % | Protokoll oder Quelle |
| 19 | PaLM 540B Unreleased | Google Research | 82,3 % | – |
| 20 | Mistral-7B-Instruct-v0.1 Open weights (unrestricted) | Mistral AI | 82,2 % | – |
| 21 | Mistral-7B-Instruct-v0.2 Open weights (unrestricted) | Mistral AI | 82,2 % | – |
| 22 | Megatron-Turing NLG 530B Unreleased | Microsoft,Nvidia | 82 % | Protokoll oder Quelle |
| 23 | Llama-2-34b Unreleased | Meta AI | 81,9 % | – |
| 24 | mpt-30b Open weights (unrestricted) | MosaicML | 81,9 % | – |
| 25 | Chinchilla (70B) Unreleased | DeepMind | 81,8 % | – |
| 26 | Gopher (280B) Unreleased | DeepMind | 81,8 % | Protokoll oder Quelle |
| 27 | Llama-3.1-8B-Instruct Open weights (restricted use) | Meta AI | 81,2 % | – |
| 28 | gemma-7b Open weights (restricted use) | Google DeepMind | 81,2 % | – |
| 29 | Phi-3.5-mini-instruct Open weights (unrestricted) | Microsoft | 81 % | – |
| 30 | mpt-7b Open weights (unrestricted) | MosaicML | 80,6 % | Protokoll oder Quelle |
| 31 | Llama-2-13b Open weights (restricted use) | Meta AI | 80,5 % | – |
| 32 | PaLM 62B | – | 80,5 % | – |
| 33 | text-davinci-001 API access | OpenAI | 80,4 % | – |
| 34 | internlm-20b | – | 80,3 % | Protokoll oder Quelle |
| 35 | LLaMA-13B Open weights (non-commercial) | Meta AI | 80,1 % | Protokoll oder Quelle |
| 36 | Qwen-14B Open weights (restricted use) | Alibaba | 79,9 % | Protokoll oder Quelle |
| 37 | LLaMA-7B Open weights (non-commercial) | Meta AI | 79,8 % | Protokoll oder Quelle |
| 38 | Llama-2-7b Open weights (restricted use) | Meta AI | 78,8 % | – |
| 39 | Baichuan-2-13B-Base Open weights (restricted use) | Baichuan | 78,1 % | Protokoll oder Quelle |
| 40 | Qwen-7B Open weights (restricted use) | Alibaba | 77,9 % | Protokoll oder Quelle |
| 41 | internlm-7b | – | 77,9 % | Protokoll oder Quelle |
| 42 | vicuna-13b-v1.1 | – | 77,4 % | – |
| 43 | gemma-2b Open weights (restricted use) | Google DeepMind | 77,3 % | – |
| 44 | RedPajama-INCITE-7B-Base | – | 76,9 % | – |
| 45 | falcon-7b Open weights (unrestricted) | Technology Innovation Institute | 76,7 % | Protokoll oder Quelle |
| 46 | gpt-neox-20b Open weights (unrestricted) | EleutherAI | 76,7 % | – |
| 47 | Baichuan-7B Open weights (non-commercial) | Baichuan | 76,2 % | Protokoll oder Quelle |
| 48 | open_llama_7b | – | 76 % | – |
| 49 | opt-13b | – | 75,7 % | – |
| 50 | xgen-7b-8k-base Open weights (unrestricted) | Salesforce | 75,5 % | – |
| 51 | dolly-v2-12b Open weights (unrestricted) | Databricks | 75,4 % | – |
| 52 | gpt-j-6b Open weights (unrestricted) | EleutherAI,LAION | 75,4 % | – |
| 53 | Cerebras-GPT-13B Open weights (unrestricted) | Cerebras Systems | 73,5 % | – |
| 54 | Mistral-7B-v0.1 Open weights (unrestricted) | Mistral AI | 73,4 % | – |
| 55 | Qwen-1.8B | – | 73,3 % | Protokoll oder Quelle |
| 56 | gpt-neo-2.7B Open weights (unrestricted) | EleutherAI | 72,9 % | – |
| 57 | gpt2-xl Open weights (unrestricted) | OpenAI | 70,5 % | – |
| 58 | chatglm2-6b | – | 69,6 % | Protokoll oder Quelle |
| 59 | opt-1.3b Open weights (non-commercial) | Meta AI | 69 % | – |
| 60 | stablelm-tuned-alpha-7b | – | 65,8 % | – |
