other
0.149B parameters
Commercial OK
Reviewed May 2026

GTE ModernBERT Base

GTE ModernBERT Base is a 149M-parameter English embedder built on AnswerDotAI's ModernBERT backbone, producing 768-dim vectors with native 8192-token context via alternating local/global attention. It pairs ModernBERT's modern architecture (RoPE, GeGLU, unpadding) with Alibaba's GTE retrieval pretraining, yielding an MTEB score of ~64 at less than half the parameters of BGE-large.

License: apache-2.0·Context: 8,192 tokens
BLK · VERDICT

Our verdict

OP · Fredoline Eruo|VERIFIED MAY 29, 2026
unrated

The most architecturally interesting English embedder of 2025 — ModernBERT's design choices (RoPE + unpadding + alternating attention) translate directly into throughput wins at no quality cost. The clear successor to nomic-v1.5 for English-only edge deployments. Skip if you need Matryoshka truncation or multilingual coverage.

Overview

GTE ModernBERT Base is a 149M-parameter English embedder built on AnswerDotAI's ModernBERT backbone, producing 768-dim vectors with native 8192-token context via alternating local/global attention. It pairs ModernBERT's modern architecture (RoPE, GeGLU, unpadding) with Alibaba's GTE retrieval pretraining, yielding an MTEB score of ~64 at less than half the parameters of BGE-large.

Strengths

  • 149M params hitting MTEB ~64 — best quality-per-param in the English embedder tier
  • ModernBERT alternating local/global attention gives true 8K context at near-flash-attention speeds
  • Apache-2.0 with documented training mix — production-safe
  • Unpadded inference via Flash Attention 2 — 2-4x faster than BERT-base at the same batch size

Weaknesses

  • 768-dim only, no Matryoshka truncation support
  • English-only — no multilingual ModernBERT-GTE variant released
  • Requires modernbert-compatible transformers (>=4.48) and ideally Flash Attention 2 for full speedup
  • Newer architecture means fewer downstream fine-tunes and ecosystem tools than BGE/E5

Quantization variants

Each quantization trades model quality for file size and VRAM. Q4_K_M is the most popular starting point.

QuantizationFile sizeVRAM required
Q4_K_M0.1 GB1 GB

Get the model

HuggingFace

Original weights

huggingface.co/Alibaba-NLP/gte-modernbert-base

Source repository — direct quantization required.

Hardware that runs this

Cards with enough VRAM for at least one quantization of GTE ModernBERT Base.

Compare alternatives

Models worth comparing

Same parameter band, plus what's one tier above and below — so you can decide what actually fits your hardware.

Step down
Smaller — faster, runs on weaker hardware
No verdicted models in the next tier down yet.

Frequently asked

What's the minimum VRAM to run GTE ModernBERT Base?

1GB of VRAM is enough to run GTE ModernBERT Base at the Q4_K_M quantization (file size 0.1 GB). Higher-quality quantizations need more.

Can I use GTE ModernBERT Base commercially?

Yes — GTE ModernBERT Base ships under the apache-2.0, which permits commercial use. Always read the license text before deployment.

What's the context length of GTE ModernBERT Base?

GTE ModernBERT Base supports a context window of 8,192 tokens (about 8K).

Source: huggingface.co/Alibaba-NLP/gte-modernbert-base

Reviewed by RunLocalAI Editorial. See our editorial policy for how we research and verify model claims.

Related — keep moving

Before you buy

Verify GTE ModernBERT Base runs on your specific hardware before committing money.