/ LLM-Benchmarks / Agentic Coding
Agentic Coding: welches Modell mehrstufige Code-Arbeit schafft
Mehrstufige Code-Bearbeitung und Tool-Nutzung in agentischen Workflows, aus LiveBench.
DatenLiveBenchRelease 2026-06-25abgerufen
LiveBench Agentic Coding · CC BY-SA 4.0
Originalquelle ansehen| # | Modell | Wertung | Kosten pro gelöster Aufgabe |
|---|---|---|---|
| 1 | DeepSeek V4.1 Flash Max EffortDeepSeek · Open Weights | 77.3% | $0.029 |
| 2 | Claude 5.5 Opus Thinking Max EffortAnthropic | 71.7% | $0.80 |
| 3 | Claude Fable 5.1 Max EffortAnthropic | 66.1% | $1.21 |
| 4 | Claude 5 Opus Thinking Max EffortAnthropic | 65.2% | $0.70 |
| 5 | DeepSeek V4 Flash Vision ExpDeepSeek · Open Weights | 65.1% | $0.051 |
| 6 | Qwen 3.8 MaxAlibaba · Open Weights | 64.6% | $0.28 |
| 7 | Muse Spark 1.3 xHigh EffortMeta | 64.1% | $0.22 |
| 8 | Claude Fable 5 Max EffortAnthropic | 62.2% | $1.44 |
| 9 | Kimi K3Moonshot AI · Open Weights | 62.2% | $0.35 |
| 10 | Qwen 3.8 Flash NextAlibaba · Open Weights | 61.6% | $0.042 |
| 11 | Qwen3.8 27BAlibaba · Open Weights | 61.4% | $0.094 |
| 12 | GLM-5.3Z.AI · Open Weights | 60.9% | $0.45 |
| 13 | Claude Sonnet 5 xHigh EffortAnthropic | 59.4% | $0.51 |
| 14 | Muse Spark 1.1 xHigh EffortMeta | 58.5% | $0.20 |
| 15 | Gemini 3.7 Flash HighGoogle | 58.3% | $0.16 |
| 16 | Muse Spark 1.2 xHigh EffortMeta | 57.6% | $0.38 |
| 17 | GPT-6 Astra Max EffortOpenAI | 57.3% | $0.74 |
| 18 | Grok 4.6 xHighxAI | 57.0% | $0.21 |
| 19 | GLM-5.3 FlashZ.AI · Open Weights | 56.8% | $0.031 |
| 20 | Grok 4.5xAI | 56.5% | $0.13 |
| 21 | GPT-5.6 Sol Max EffortOpenAI | 56.2% | $0.52 |
| 22 | GPT-5.6 Terra Max EffortOpenAI | 54.9% | $0.35 |
| 23 | DeepSeek V4 Pro 0813DeepSeek · Open Weights | 54.9% | $0.044 |
| 24 | GPT-6.1 Sol Max EffortOpenAI | 54.5% | $0.14 |
| 25 | Gemini 3.8 Flash HighGoogle | 54.2% | $0.31 |
| 26 | GPT-5.5 Thinking xHigh EffortOpenAI | 54.0% | $0.43 |
| 27 | Grok 4.7 xHighxAI | 54.0% | $0.72 |
| 28 | GPT-5.4 Thinking xHigh EffortOpenAI | 53.8% | $0.39 |
| 29 | GPT-6 Sol Max EffortOpenAI | 52.9% | $0.27 |
| 30 | Ox Alpha MaxStealth | 52.6% | – |
| 31 | GLM-5.2Z.AI · Open Weights | 51.8% | $0.23 |
| 32 | GPT-6 Luna Max EffortOpenAI | 51.2% | $0.026 |
| 33 | Claude 4.7 Opus Thinking xHigh EffortAnthropic | 50.7% | $0.53 |
| 34 | Claude 4.8 Opus Thinking Max EffortAnthropic | 50.5% | $0.98 |
| 35 | GPT-5.2 HighOpenAI | 50.3% | $0.23 |
| 36 | GPT-5.2 CodexOpenAI | 49.4% | $0.19 |
| 37 | Inkling xHigh EffortThinking Machines · Open Weights | 49.4% | $0.31 |
| 38 | Gemini 3.5 Flash HighGoogle | 49.0% | $0.25 |
| 39 | Claude 4.6 Opus Thinking High EffortAnthropic | 49.0% | $0.40 |
| 40 | GPT-5.6 Luna Max EffortOpenAI | 48.4% | $0.17 |
| 41 | Kimi K2.6 ThinkingMoonshot AI · Open Weights | 46.9% | $0.17 |
| 42 | DeepSeek V4 Flash 0731DeepSeek · Open Weights | 46.8% | $0.060 |
| 43 | GPT-5.4 Nano xHighOpenAI | 46.8% | $0.091 |
| 44 | Grok Build 0.1xAI | 45.8% | $0.024 |
| 45 | Kimi K2.7 CodeMoonshot AI · Open Weights | 45.7% | $0.10 |
| 46 | Gemini 3.5 Flash-Lite HighGoogle | 45.3% | $0.069 |
| 47 | Gemini 3.1 Pro Preview HighGoogle | 44.1% | $0.29 |
| 48 | Qwen 3.7 MaxAlibaba | 43.6% | $0.18 |
| 49 | Gemini 3.6 Flash HighGoogle | 43.4% | $0.23 |
| 50 | Claude 4.6 Sonnet Thinking Medium EffortAnthropic | 42.6% | $0.31 |
| 51 | GPT-5.4 Mini xHighOpenAI | 41.7% | $0.33 |
| 52 | Qwen 3.6 PlusAlibaba | 41.4% | $0.23 |
| 53 | Minimax M3MiniMax | 40.7% | $0.060 |
| 54 | Claude 4.5 Opus Thinking High EffortAnthropic | 39.7% | $0.61 |
| 55 | Claude Sonnet 5.5 xHigh EffortAnthropic | 39.3% | $0.14 |
| 56 | Qwen 3.6 27BAlibaba · Open Weights | 39.3% | $0.20 |
| 57 | Nemotron 3 Ultra 550B A55BNVIDIA · Open Weights | 38.7% | $0.37 |
| 58 | Grok 4.3xAI | 18.5% | $0.061 |
Sie brauchen ein Modell für Ihren konkreten Fall?
Benchmarks bewerten Modelle anhand fremder Aufgaben. Für einen Geschäftsprozess teste ich die engere Wahl mit Ihren eigenen Daten, Kosten und Datenschutzvorgaben und empfehle eines.