Open-Weight-Modelle (Llama, Qwen, DeepSeek, Mistral…) im Vergleich zu den proprietären Modellen (GPT, Claude, Gemini, Grok) — Live-Preise, Kontext, Benchmarks und Fähigkeiten. Keine Werbung — nur die Fakten.
| # | Model | Creator | Intelligence | Coding |
|---|---|---|---|---|
| 1 | Claude Opus 5.5 (Max, Default Fallback) | Anthropic | 58.4 | — |
| 2 | GPT-6 Astra (Max) | OpenAI | 56.9 | 76.9 |
| 3 | GPT-6.1 Sol (Max) | OpenAI | 54.5 | — |
| 4 | Claude Sonnet 5.5 (Max, Default Fallback) | Anthropic | 54.2 | — |
| 5 | GPT-6.1 Terra (Max) | OpenAI | 51.0 | — |
| 6 | Gemini 4 Argon (High) | 50.4 | — | |
| 7 | Claude Fable 5.1 (Xhigh, Default Fallback) | Anthropic | 49.7 | 80.7 |
| 8 | Muse Wonder Full (Max) | Meta | 49.7 | — |
| 9 | Kimi K3.1 Pro | Kimi | 46.0 | — |
| 10 | Muse Wonder (Max) | Meta | 43.3 | — |
| 11 | Muse Spark 1.3 (Max) | Meta | 36.4 | 75.8 |
| 12 | Claude Haiku 5.5 (Max) | Anthropic | 36.2 | — |
| 13 | Grok 4.7 (High) | SpaceXAI | 35.0 | — |
| 14 | Qwen3.8 Max (0902) | Alibaba | 33.6 | 76.2 |
| 15 | DeepSeek V4.1 Flash 1017 (Max) | DeepSeek | 32.5 | — |
| 16 | GLM-5.3 (Max) | Z AI | 31.7 | 74.8 |
| 17 | Gemini 3.8 Flash (High) | 30.5 | 76.3 | |
| 18 | MiMo-V2.6-Pro | Xiaomi | 29.9 | — |
| 19 | Ling 3.1 Flash 1019 | InclusionAI | 29.5 | — |
| 20 | MiniMax-M3.1-Flash | MiniMax | 28.9 | — |
| 21 | Step 5 Preview | StepFun | 28.5 | — |
| 22 | GPT-6 Luna (Max) | OpenAI | 27.8 | — |
| 23 | MiMo-V2.6-Flash | Xiaomi | 26.7 | — |
| 24 | Mistral Large 4 Preview | Mistral | 25.3 | — |
| 25 | Muse Spark Open (Xhigh) | Meta | 24.7 | — |
| 26 | DeepSeek V4 Pro 0813 (Max) | DeepSeek | 23.1 | 68.8 |
| 27 | Qwen3.8 27B (Xhigh) | Alibaba | 20.1 | 68.1 |
| 28 | MiniMax-M3 | MiniMax | 18.7 | 58.6 |
| 29 | K2 Horizon 375B A23B | Institute of Foundation Models | 17.8 | 61.5 |
| 30 | Beam 1.0 | Reflection | 17.7 | — |
| 31 | Nemotron 3.5 Super VL | NVIDIA | 13.6 | — |
| 32 | Inkling Small | Thinking Machines | 12.9 | 52.9 |
| 33 | Inkling (Xhigh) | Thinking Machines | 12.5 | 52.1 |
| 34 | Nemotron 3 Ultra 550B A55B (Reasoning) | NVIDIA | 11.8 | 49.3 |
| 35 | Gemini 3.5 Flash-Lite | 11.7 | 49.3 | |
| 36 | Muse Glimmer (High) | Meta | 6.3 | 49.0 |
| 37 | Command A+ | Cohere | 6.2 | 27.8 |
| 38 | NVIDIA Nemotron 3 Nano 4B | NVIDIA | 1.8 | 8.0 |
| 39 | A.X-K2 | SK Telecom | — | 38.8 |
| 40 | Agnes 2.5 Pro Beta | Sapiens AI | — | 62.3 |
| 41 | Apertus 70B Instruct | Swiss AI Initiative | — | — |
| 42 | Apertus 8B Instruct | Swiss AI Initiative | — | — |
| 43 | Apodex 1.1 | Apodex | — | 60.8 |
| 44 | Apriel-v1.5-15B-Thinker | ServiceNow | — | — |
| 45 | Apriel-v1.6-15B-Thinker | ServiceNow | — | — |
| 46 | Celeris-1 | Celeris | — | 14.4 |
| 47 | Claude 2.0 | Anthropic | — | 12.9 |
| 48 | Claude 2.1 | Anthropic | — | 14.0 |
| 49 | Claude 3 Haiku | Anthropic | — | — |
| 50 | Claude 3 Opus | Anthropic | — | 19.5 |
| 51 | Claude 3 Sonnet | Anthropic | — | — |
| 52 | Claude 3.5 Haiku | Anthropic | — | 15.9 |
| 53 | Claude 3.5 Sonnet (June '24) | Anthropic | — | 26.0 |
| 54 | Claude 3.7 Sonnet (Non-reasoning) | Anthropic | — | — |
| 55 | Claude 4 Opus (Non-reasoning) | Anthropic | — | — |
| 56 | Claude 4 Sonnet (Non-reasoning) | Anthropic | — | — |
| 57 | Claude 4.1 Opus (Reasoning) | Anthropic | — | — |
| 58 | Claude 4.5 Haiku (Non-reasoning) | Anthropic | — | — |
| 59 | Claude 4.5 Sonnet (Non-reasoning) | Anthropic | — | — |
| 60 | Claude Fable 5 (Max, Opus 4.8 Fallback) | Anthropic | — | 76.5 |
So wird das eingestuft: Modelle sind nach der Beliebtheit der realen Nutzung (wöchentliche Tokens, die über Apps verarbeitet werden) über die OpenRouter API geordnet. Nur open-weight sind enthalten (Llama, DeepSeek, Qwen, Mistral, Gemma, Kimi, GLM, Phi, Nemotron und mehr). Die Preise gelten pro Million Tokens; der Kontext ist das maximale Fenster. Bewegungs-Pfeile vergleichen den heutigen Rang mit dem vorherigen Snapshot. Daten werden täglich aktualisiert.