Evals · conare-embed-1

LMEB, memory retrieval

Finding the right moment in long conversations and personal records. The 13 tasks whose data allows commercial use.

conare-embed-1

75.0

nDCG@10, higher is better · #1 of 106

  1. 1conare-embed-175.0
  2. 2nvidia/Nemotron-3-Embed-8B72.5
  3. 3BAAI/bge-multilingual-gemma268.7
  4. 4nvidia/Nemotron-3-Embed-1B68.7
  5. 5tencent/KaLM-Embedding-Gemma3-12B68.4
  6. 6nvidia/NV-Embed-v266.4
  7. 7codefuse-ai/F2LLM-v2-8B66.2
  8. 9microsoft/harrier-oss-v1-0.6b65.7
  9. 10NovaSearch/stella_en_400M_v565.6
  10. 20perplexity-ai/pplx-embed-v1-0.6b64.2
  11. 84mixedbread-ai/mxbai-embed-large-v156.6

nDCG@10, higher is better · read 14 October 2026

One of the 13 tasks, LongMemEval, was used to choose this model. Without it: 73.7 against Nemotron-3-Embed-8B's 71.4. Not a leaderboard submission.

How it was measured

LMEB, the 13 of its 22 tasks whose datasets carry a licence that permits commercial use, test split, nDCG@10 averaged over tasks, no task instructions. Every other row is a score published on the MTEB leaderboard for all 13 tasks (105 models, data of 12 October 2026).

See it on your own data.

Book a demo