LangChain tests Jev as an agent-eval judge against GPT-5.6 and Claude Sonnet 4.6

LangChain @LangChainOriginally posted 2026.09.22

Open original on X

We tested Jev against GPT-5.6 Luna, GPT-5.6 Terra, and Claude Sonnet 4.6 as agent-eval judges. ✅ Jev matched a human reviewer on every call, with up to 913x lower variance, at 0.44s per call vs 2.16-2.83s for the LLMs. ✅ Cost for the full run: $0.34 with Jev. $28.17 with Claude Sonnet 4.6.