Models
Price comparison
EN

Leaderboards for AI models

Which model leads at which task – with the cheapest price in the EU and worldwide, open weights and the GPU memory for self-hosting.

We only show benchmarks whose source allows republishing, and a list only from 10 models with a score and an active offer. Rank = position among the models of this catalogue with a score, not in the original list. Models for self-hosting by graphics card: Self-hosting

Knowledge

MMLU-Pro (TIGER-Lab Leaderboard)allgemein - Wissen ueber 14 Fachgebiete↑ 42 models · leader Gemini 3.1 Pro (Vorschau) (91.16 %)Source: huggingface.co SimpleQA Verified (Epoch AI)allgemein - Faktenwissen ohne Hilfsmittel, Halluzinationsneigung↑ 85 models · leader Gemini 4 Argon (76 %)Source: epoch.ai

Mathematics

FrontierMath Tiers 1-3 v2 (Epoch AI)allgemein - Mathematik bis Forschungsniveau↑ 91 models · leader GPT-6 astra (93.7 %)Source: epoch.ai OTIS Mock AIME 2024-2025 (Epoch AI)allgemein - Wettbewerbsmathematik↑ 153 models · leader Claude Fable 5.1 (100 %)Source: epoch.ai

Coding

SWE-bench Verified (Epoch AI)allgemein - Fehlerbehebung in echten Code-Repositories↑ 33 models · leader Opus 4.7 (83.5 %)Source: epoch.ai

Tool calling

Berkeley Function-Calling Leaderboard V4allgemein - Werkzeug- und Funktionsaufrufe, agentische Ablaeufe↑ 42 models · leader Opus 4.5 (77.47 %)Source: gorilla.cs.berkeley.edu

Hallucination

Vectara HHEM-2.3Zusammenfassung - Treue zur Vorlage↓ 82 models · leader GPT-5.4 nano (3.1 %)Source: github.com

Language tasks and extraction

searchit Entity ExtractionExtract details such as names, companies and amounts from German-language texts (measured by searchit).↑ 53 models · leader GPT-6 astra (77.1 %)Source: How the data is made

No leaderboard (scores on the model pages only): MÖVE (Bundesdruckerei), arXiv 2606.13111 (Licence of the source unclear); Structured Output Benchmark (SOB), arXiv 2604.25359 (Licence of the source unclear).

General

No leaderboard (scores on the model pages only): LiveBench (Aufgabensatz 2026-01-08) (Licence of the source unclear).

Image understanding

No leaderboard (scores on the model pages only): HallusionBench (llm-stats-Abzug) (Licence of the source unclear); MMMU-Pro (llm-stats-Abzug, Rang 1..50 von 69) (Licence of the source unclear); OCRBench v2, englische Spur (llm-stats-Abzug) (Licence of the source unclear).

Embedding

MTEB ArguAnaRetrieval (en)↑ 38 models · leader gemini-embedding-001 (86.4)Source: huggingface.co MTEB EnglishFinance1RetrievalRetrieval (en)↑ 34 models · leader Voyage 4 large (82.2)Source: huggingface.co MTEB EnglishFinance2RetrievalRetrieval (en)↑ 34 models · leader Voyage 4 large (91)Source: huggingface.co MTEB EnglishFinance3RetrievalRetrieval (en)↑ 34 models · leader gemini-embedding-001 (83.3)Source: huggingface.co MTEB EnglishFinance4RetrievalRetrieval (en)↑ 34 models · leader Voyage 4 large (62)Source: huggingface.co MTEB EnglishHealthcare1RetrievalRetrieval (en)↑ 32 models · leader Voyage 4 large (67.5)Source: huggingface.co MTEB German1RetrievalRetrieval (de)↑ 35 models · leader gemini-embedding-001 (97.6)Source: huggingface.co MTEB GermanDPRRetrieval (de)↑ 22 models · leader most-embed-de (85.6)Source: huggingface.co MTEB GermanHealthcare1RetrievalRetrieval (de)↑ 35 models · leader Voyage 4 large (91.4)Source: huggingface.co MTEB GermanLegal1RetrievalRetrieval (de)↑ 35 models · leader Voyage 4 large (75.5)Source: huggingface.co MTEB GermanQuAD-RetrievalRetrieval (de)↑ 26 models · leader most-embed-de (97.7)Source: huggingface.co MTEB LegalQuADRetrieval (de)↑ 44 models · leader most-embed-de (76.2)Source: huggingface.co MTEB NFCorpusRetrieval (en)↑ 29 models · leader Qwen3 Embedding 8B (41.4)Source: huggingface.co MTEB SciFactRetrieval (en)↑ 28 models · leader Qwen3 Embedding 8B (78.5)Source: huggingface.co MTEB TRECCOVIDRetrieval (en)↑ 37 models · leader Qwen3 Embedding 8B (95)Source: huggingface.co

No leaderboard (scores on the model pages only): Text zu Foto (Mittel ueber 3 Aufgaben) (too few models with an offer: 0, from 10); ViDoRe v1 (Mittel ueber 10 Aufgaben, Dokumentseite als Bild) (too few models with an offer: 0, from 10); ViDoRe v3 (Mittel ueber 9 Aufgaben) (too few models with an offer: 1, from 10).

Other

No leaderboard (scores on the model pages only): ExtractBench: A Benchmark for Schema-Guided Enterprise Document Extraction (arXiv:2607.29677) (Licence of the source unclear); LLMStructBench: Benchmarking Large Language Model Structured Data Extraction (arXiv:2602.14743) (Licence of the source unclear).