Reason-ModernColBERT
A 149M late-interaction retriever that out-reasons models 45x its size on BRIGHT.
About
Reason-ModernColBERT is LightOn's answer to the question of whether small models can do reasoning retrieval. It takes the GTE-ModernColBERT backbone (149M parameters) and fine-tunes it on ReasonIR-HQ, the synthetic reasoning dataset from Meta's ReasonIR project, using LightOn's PyLate library for the ColBERT-style late-interaction training.
The model card claims it outperforms all existing retrieval models up to 7B parameters on the BRIGHT benchmark and even improves on ReasonIR-8B itself by more than 2.5 nDCG@10 on average across the Stack Exchange splits, attributing the win to late interaction's token-level matching. With over 62,000 Hugging Face downloads it is one of the most adopted reasoning retrievers, but note the CC-BY-NC-4.0 license inherited from the distillation data: research use only.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
- BRIGHT: outperforms all models up to 7B (45x its size); +2.5 nDCG@10 avg over ReasonIR-8B on Stack Exchange splits (provider-reported, model card)
- No independent MTEB/BEIR replication published
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (verified tier, metrics-v1). Evidence: HF API confirms CC-BY-NC-4.0 license, May 2025 release, and 62K+ downloads; training recipe (ReasonIR-HQ + PyLate) is public and reproducible. Strengths: remarkable efficiency story, real adoption, transparent training. Docked for: (1) NC license bars commercial use; (2) headline BRIGHT figures are provider-reported with no independent replication; (3) distillation-based training means its reasoning ability is borrowed, not learned from scratch.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.0/10
- Maintenance · recency, releases, issue load
- 7.0/10
- Openness · license + docs
- 7.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · HF API confirms CC-BY-NC-4.0, May 2025 release, 62K+ downloads
- · 149M ModernBERT late-interaction model trained on ReasonIR-HQ via PyLate; recipe is public
- · Headline BRIGHT figures are provider-reported with no independent replication