LLM Evaluation Framework。EleutherAI lm-evaluation-harness 0.4 (50+ Tasks・Standard)・OpenAI evals・RAGAS (RAG Eval・Faithfulness/Answer Relevancy)・promptfoo 0.106 (Web UI/CI)・DeepEval (Confident-AI)・Inspect AI (UK AISI)・METR Time Horizon・LLMonitor・Phoenix (Arize)・LangSmith Eval・Braintrust・¥0 OSS-¥¥¥¥¥/月、2026年Inspect AI+RAGAS主流。