AI Hub
All models
LLMOpen weights

RoBERTa

Meta

Updated May 21, 2026 · Benchmarks via Artificial Analysis, specs & pricing via OpenRouter · Methodology · Report an error

Specifications

Type
LLM
Access
Open weights
Released
July 26, 2019
Parameters
355M
Context window
512 tokens
Input
Text
Output
Text

Our take

Showed BERT was undertrained: with more data and tuning, the same architecture got substantially better. A lesson in the value of training recipe over novelty.

Related research

Compare RoBERTa with

See all RoBERTa alternatives →