| Modellslot | Typ | Version | Overall | Coding | Reasoning | Tool/Agent | Long/RAG | Multi | Local | Gap | Trend | Quellen |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
Claude Opus Anthropic | Cloud | Opus 4.6 | 100% | 98% | 99% | 96% | 94% | 91% | 0% | 0 | → | AA, SWE, BFCL |
GPT Flagship OpenAI | Cloud | GPT-5.5 | 98% | 96% | 100% | 97% | 91% | 95% | 0% | -2 | +1 | AA, LiveBench |
Gemini Pro Google | Cloud | Gemini 3 Pro | 94% | 89% | 96% | 89% | 100% | 98% | 0% | -6 | → | AA, RULER, MMMU |
Claude Sonnet Anthropic | Cloud | Sonnet 4.6 | 91% | 95% | 88% | 90% | 86% | 84% | 0% | -9 | → | SWE, BFCL |
Qwen Coder Alibaba / Open Weights | Open | Qwen Coder 3.x | 82% | 89% | 73% | 70% | 78% | 42% | 74% | -18 | +3 | Aider, HF, Ollama |
DeepSeek Reasoner DeepSeek / Open Weights | Open/API | DeepSeek R2 | 80% | 76% | 88% | 66% | 74% | 30% | 38% | -20 | +2 | LiveBench, GPQA |
Llama Instruct Meta / Open Weights | Open | Llama 4.x | 74% | 70% | 76% | 64% | 82% | 62% | 83% | -26 | → | HF, RULER, MMMU |
Mistral Large Mistral | Hybrid | Large 3 | 72% | 68% | 73% | 70% | 76% | 58% | 52% | -28 | -1 | AA, BFCL |
Gemma Google / Open Weights | Open | Gemma 4 12B | 66% | 55% | 68% | 50% | 72% | 78% | 92% | -34 | +4 | HF, Ollama, MMMU |
Qwen Consumer Laptop/MacBook class | Local | Qwen 3 8B GGUF | 58% | 52% | 61% | 44% | 67% | 24% | 96% | -42 | +2 | HF, GGUF, Ollama |