Is papr-embed-code-1-0.6b state of the art?Pending benchmarks.Announced Oct 28, 2026; no independent results yet.

The details so far

Lab

Papr

Announced

Oct 28, 2026

API

Live · papr-embed-code-1-0.6b · POST https://memory.papr.ai/v1/embeddings with model papr-embed-code-1-0.6b

Open weights

Published · Hugging Face papr-ai/papr-embed-code-1-0.6b, fp32, Apache-2.0 (repo created Oct 19, card added Oct 21)

Parameters

0.6B (fine-tune of Qwen3-Embedding-0.6B)

Task

Text embeddings for code retrieval; one dense vector per input, no schema input

Base model

Qwen/Qwen3-Embedding-0.6B

Embedding size

1,024 dimensions; last-token pooling, L2-normalized; up to 8,192 tokens

Languages

English and code only

Training data note

Trained on the train splits of all 10 CoIR tasks and SciFact, so CoIR and SciFact are in-domain, per Papr

Tags marked unsure are not stated on the lab's own pages.

What would let us call it

Verdicts use only independent boards we already track. Two consecutive snapshots are required before any crown.

• Publicly announced with a source
• Usable by the public (API or weights)
• Listed on an independent board we track
• Enough results for a rank in a category
• Holds across two snapshots (crowns only)

What the lab claims

Self-reported. Shown for context; not used in any verdict.

Lab claim: CoIR mean nDCG@10 of 77.19 over 10 tasks (in-domain, per Papr), run with stock mteb 2.10 in fp32.

Lab claim: on held-out retrieval tasks it trails Qwen3-Embedding-0.6B on ArguAna, SciDocs and TREC-COVID (82.70 vs 89.30) and gains on NFCorpus, FiQA and CodeEditSearch.

Sources