OpenBookQA
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 42 Ergebnissen; die Tabelle führt alle auf.
| # | Modell | Organisation | Ergebnis | Nachweis |
|---|---|---|---|---|
| 1 | Phi-3-mini-4k-instruct Open weights (unrestricted) | Microsoft | 88 % | – |
| 2 | Phi-3-small-8k-instruct Open weights (unrestricted) | Microsoft | 88 % | – |
| 3 | Phi-3-medium-128k-instruct Open weights (unrestricted) | Microsoft | 87,4 % | – |
| 4 | GPT-3.5 Turbo (Nov 2023) API access | OpenAI | 86 % | – |
| 5 | Mixtral-8x7B-v0.1 Open weights (unrestricted) | Mistral AI | 85,8 % | – |
| 6 | Meta-Llama-3-8B-Instruct Open weights (restricted use) | Meta AI | 82,6 % | – |
| 7 | Mistral-7B-v0.1 Open weights (unrestricted) | Mistral AI | 79,8 % | – |
| 8 | gemma-7b Open weights (restricted use) | Google DeepMind | 78,6 % | – |
| 9 | phi-2 Open weights (unrestricted) | Microsoft | 73,6 % | – |
| 10 | PaLM 540B Unreleased | Google Research | 68 % | Protokoll oder Quelle |
| 11 | text-davinci-001 API access | OpenAI | 65,4 % | – |
| 12 | falcon-180B Open weights (restricted use) | Technology Innovation Institute | 64,2 % | – |
| 13 | GLaM (MoE) Unreleased | 63 % | – | |
| 14 | LLaMA-65B Open weights (non-commercial) | Meta AI | 60,2 % | – |
| 15 | Llama-2-70b-hf Open weights (restricted use) | Meta AI | 60,2 % | – |
| 16 | LLaMA-33B Open weights (non-commercial) | Meta AI | 58,6 % | – |
| 17 | Llama-2-7b Open weights (restricted use) | Meta AI | 58,6 % | – |
| 18 | Llama-2-34b Unreleased | Meta AI | 58,2 % | – |
| 19 | PaLM 2-M | – | 57,4 % | Protokoll oder Quelle |
| 20 | LLaMA-7B Open weights (non-commercial) | Meta AI | 57,2 % | – |
| 21 | Llama-2-13b Open weights (restricted use) | Meta AI | 57 % | – |
| 22 | falcon-40b Open weights (unrestricted) | Technology Innovation Institute | 56,6 % | – |
| 23 | LLaMA-13B Open weights (non-commercial) | Meta AI | 56,4 % | – |
| 24 | PaLM 2-S | – | 56,2 % | Protokoll oder Quelle |
| 25 | mpt-30b Open weights (unrestricted) | MosaicML | 52 % | – |
| 26 | falcon-7b Open weights (unrestricted) | Technology Innovation Institute | 51,6 % | – |
| 27 | mpt-7b Open weights (unrestricted) | MosaicML | 51,4 % | – |
| 28 | PaLM 62B | – | 50,4 % | – |
| 29 | xgen-7b-8k-base Open weights (unrestricted) | Salesforce | 40,2 % | – |
| 30 | RedPajama-INCITE-7B-Base | – | 40 % | – |
| 31 | opt-13b | – | 39,8 % | – |
| 32 | dolly-v2-12b Open weights (unrestricted) | Databricks | 39,2 % | – |
| 33 | open_llama_7b | – | 39 % | – |
| 34 | gpt-neox-20b Open weights (unrestricted) | EleutherAI | 38,8 % | – |
| 35 | gpt-j-6b Open weights (unrestricted) | EleutherAI,LAION | 38,2 % | – |
| 36 | phi-1_5 Open weights (unrestricted) | Microsoft | 37,2 % | – |
| 37 | Cerebras-GPT-13B Open weights (unrestricted) | Cerebras Systems | 35,8 % | – |
| 38 | vicuna-13b-v1.1 | – | 33 % | – |
| 39 | stablelm-tuned-alpha-7b | – | 32,4 % | – |
| 40 | opt-1.3b Open weights (non-commercial) | Meta AI | 24 % | – |
| 41 | gpt-neo-2.7B Open weights (unrestricted) | EleutherAI | 23,2 % | – |
| 42 | gpt2-xl Open weights (unrestricted) | OpenAI | 22,4 % | – |
