Harrison Chase @hwchase17: Nemotron 3 Ultra hit 86%. Claude Opus hit 87%. At one-tenth the cost. Chase runs LangChain and disclosed it on st…

Harrison Chase @hwchase17: Nemotron 3 Ultra hit 86%. Claude Opus hit 87%. At one-tenth the cost. Chase runs LangChain and disclosed it on stage: inside LangChain's internal deep-agents benchmark, open-weight Nemotron 3 Ultra scored 86% to Opus 4.8's 87%, with DeepSeek and Minimax at 82-83%. Nemotron costs roughly 10x less, and you can download the weights. One point of separation, at a tenth of the price, on a model you host yourself. Note the sourcing: Chase is a Nemotron coalition member and the benchmark is self-reported. Gavin Baker @GavinSBaker of Atreides pushes the other way - his case is that compute scaling keeps the frontier moat wide, and near-parity benchmarks understate the real lead. If the gap holds on a neutral benchmark, Anthropic and OpenAI lose pricing power on agentic workloads. Watch for third-party replication. The full benchmark breakdown and what it means for closed-model pricing: https://podcastalpha.substack.com/p/jensen-huang-own-your-intelligence Source: LangChain - https://www.youtube.com/watch?v=Yy3JH6dDugc Media

Source: Harrison Chase (X) | 2026-07-09