HindsightFrom the builder
We Got Early Access to NVIDIA Nemotron 3.5 Embed. It's Hindsight's New Top Embedding Model.
Agent memory project Hindsight benchmarked Nvidia's Nemotron 3.5 Embed against seven other embedding models, including OpenAI text-embedding-3-small, BGE-M3 and Qwen3-Embedding-0.6B. Hindsight says Nemotron ranked first on LoCoMo and BEAM 100K memory retrieval with 93.8 and 65.9 percent. Its 32,768-token context lets it embed whole conversation passages, and it posted 38.0 percent recall@10 on the 10-million-token BEAM tier and a 55.1 mean nDCG@10 across 27 tasks in eight languages. It took about 30 milliseconds per query on one Nvidia L4, faster than Qwen3-Embedding-0.6B, a model roughly half its size.
Nemotron finished first on memory retrieval and first on the leaderboard, and it answered queries faster than a model just over half its size.
Saved . A shortened excerpt from the official blog; the original publisher retains authorship.