CodeT5
Identifier-aware encoder-decoder model for code understanding and generation.
About
CodeT5, presented at EMNLP 2021, is a unified pretrained encoder-decoder Transformer for programming languages. Its signature contribution is an identifier-aware pretraining task that teaches the model to distinguish developer-assigned identifiers from other code tokens and to recover masked identifiers, plus a bimodal dual-generation task aligning code with natural-language comments.
It significantly outperformed prior methods on code defect detection, clone detection, and generation tasks in both directions between code and natural language. The code and pretrained models are released under the BSD-3-Clause license.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
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 (metrics tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 82; Trending score 70/100 from researcher assessment.