CodeT5
github.com
Salesforce's open-source code understanding and generation model.
Quick Information
- Category:
- Coding
- Pricing:
- Free
- Platforms:
- Web
- Free Plan:
- Yes
- Last Verified:
- September 2026
What Is CodeT5?
CodeT5 is Salesforce Research's open-source, identifier-aware encoder-decoder model for code understanding and generation, built on a similar architecture to Google's T5 framework but purpose-tuned for code. Its successor, CodeT5+, expanded into a family of models ranging from 220 million to 16 billion parameters that can flexibly operate as encoder-only, decoder-only, or full encoder-decoder — the version trained on 51.5 billion tokens reached 35.0% pass@1 on the HumanEval benchmark, outperforming OpenAI's code-cushman-001 model while remaining fully Apache 2.0 licensed.
Salesforce archived the official GitHub repository in May 2025, meaning it's no longer under active first-party development — but the model weights remain accessible on Hugging Face (CodeT5-base recorded over 22,000 monthly downloads as of 2026), and community forks continue active development and maintenance. That makes it a genuinely useful foundational model for research and lightweight applications, even though it's not a maintained commercial product in the way newer coding assistants are.
It's aimed at researchers and developers wanting an open, permissively-licensed foundational code model to build on, rather than an actively maintained commercial coding assistant — for a maintained product experience, see tools like CodeGeeX or Cline instead.
What Can You Use CodeT5 For?
Code understanding research
Provides an open, permissively-licensed model for studying code understanding tasks.
Custom code model fine-tuning
Serves as a foundational model for fine-tuning custom code generation or understanding tools.
Lightweight code generation applications
Powers lightweight applications needing code generation without a commercial API dependency.
CodeT5 Features
- Encoder-decoder architecture — Built on a T5-style architecture tuned specifically for code understanding and generation.
- Flexible model modes — CodeT5+ operates as encoder-only, decoder-only, or full encoder-decoder depending on the task.
- Apache 2.0 license — Fully open-source and permissively licensed for commercial and research use.
- Range of model sizes — CodeT5+ family spans from 220M to 16B parameters.
Who Is CodeT5 For?
- Researchers — Gets an open, permissively-licensed foundational model for code understanding research.
- Developers building custom tools — Fine-tunes CodeT5 as a base for lightweight, self-hosted code generation applications.
CodeT5 Pros & Limitations
What We Like
- Fully open-source under the permissive Apache 2.0 license
- Historically outperformed OpenAI’s code-cushman-001 on HumanEval despite being open-source
- Community forks keep the project alive despite Salesforce archiving the official repository
Limitations
- Salesforce archived the official GitHub repository in May 2025 — no longer under active first-party development
- Smaller and less capable than current frontier commercial coding models
- Requires more technical setup than a ready-to-use commercial coding assistant
CodeT5 Pricing
Free
Pricing was last verified September 2026. Check the official website for the latest pricing.
AI Listing Jungle Ranking
- Global Rank:
- #443
- Overall Score:
- 48.7/100
- Monthly Users:
- –
- Rank Change:
- ▲ 23
This ranking reflects AI Listing Jungle's own independent scoring methodology and is not an endorsement by CodeT5 or its company. How Our Rankings Work
Frequently Asked Questions
Is CodeT5 still maintained?
Salesforce archived the official GitHub repository in May 2025, but model weights remain on Hugging Face and community forks continue development.
Is CodeT5 free to use?
Yes, it’s fully open-source under the Apache 2.0 license.
How large are CodeT5+ models?
The CodeT5+ family ranges from 220 million to 16 billion parameters.
Is CodeT5 a ready-to-use coding assistant?
No, it’s a foundational model best suited to researchers and developers building custom tools, not a polished commercial product.
Our Take
CodeT5 remains a genuinely significant open-source contribution to code intelligence research, even though Salesforce's archiving of the repository in 2025 means it's no longer a first-party maintained project.
For a ready-to-use assistant rather than a research foundation, a maintained tool like CodeGeeX is the better practical choice.
CodeT5 Alternatives
CodeT5 Comparisons
Ready to try CodeT5?
Visit the official website to get started.
Visit Official WebsiteAI Listing Jungle is an independent AI tools directory. Product names, logos, and trademarks belong to their respective owners.