docs: add external-baseline comparison (bge-m3 / e5-large / embeddinggemma)

README.md

@@ -140,6 +140,19 @@ Trained for 3 epochs (early-stopped) with a cosine schedule; training loss decre
 
 **v4 is the released model.** Fine-tuning lifts in-domain retrieval by **~6%** over the base `Qwen3-Embedding-4B`, with the largest gains on top-rank precision (Accuracy@1, Recall@1). It also surpasses the 0.6B sibling (test NDCG@10 0.6695) by **+0.021 (+3.2%)** — a modest scale gain at ~7× the parameters, so the 0.6B remains the better pick for latency-sensitive serving.
 
+### Comparison with public encoders
+On the *same* in-domain test set, public multilingual SOTA encoders (each run with its own native prompt format) all fall **well below this model** — domain fine-tuning beats general-purpose scale:
+
+| Model | Params | NDCG@10 | MRR@10 | Recall@10 | Accuracy@1 |
+|-------|:---:|:---:|:---:|:---:|:---:|
+| google/embeddinggemma-300m | 0.3B | 0.6373 | 0.6664 | 0.7082 | 0.5790 |
+| BAAI/bge-m3 | 0.6B | 0.6426 | 0.6660 | 0.7261 | 0.5730 |
+| intfloat/multilingual-e5-large | 0.6B | 0.6476 | 0.6722 | 0.7313 | 0.5790 |
+| MoAI-Embedding-0.6B (sibling) | 0.6B | 0.6695 | 0.7060 | 0.7508 | 0.6190 |
+| **MoAI-Embedding-4B (this model)** | 4B | **0.6906** | **0.7283** | **0.7620** | **0.6480** |
+
+This model leads the best general-purpose baseline (e5-large) by **+0.043 NDCG@10**. _Caveat: these baselines are not tuned on BC Card data — the comparison illustrates the value of domain adaptation, not a defect in the baselines._
+
 <br>
 
 ## 2.4. Limitations