LaMDA Models by Google DeepMind
Last refreshed 2026-04-15. Next refresh: weekly.
Details
About
LaMDA, short for Language Model for Dialogue Applications, represents Google's suite of conversational large language models. Developed using the Transformer architecture first introduced by Google Research in 2017, LaMDA is finely tuned for dialogue, making it adept at understanding the nuances of open-ended conversations. This capability arises from its comprehensive training on a diverse dataset of approximately 1.56 trillion words from public dialogues and web text. The most significant iteration of LaMDA features a staggering 137 billion parameters. Google's commitment to safety and factual accuracy in LaMDA is evident through their strategies that include fine-tuning with annotated data and leveraging external knowledge sources. LaMDA is instrumental in powering Google’s products like Bard and has sparked discussions around AI sentience 235.
Decision facts
- Best fit
- coding
- Capability starting point
- LaMDA 137B
- Lowest tracked input
- Not tracked
- Closest related family
- T5Gemma
Current Variants
Use-when guidance is based on each model's tracked capabilities, context window, release date, and replacement status.
| Model | Use when | Released | Signals | Status |
|---|---|---|---|---|
| LaMDA 137B | Use when the workload needs 137B parameters. | 2021-05 | 137B parameters | Current |
| LaMDA 8B | Use when the workload needs 8B parameters. | 2021-05 | 8B parameters | Current |
| LaMDA 2B | Use when the workload needs 2B parameters. | 2021-05 | 2B parameters | Current |
Release Timeline
1 release groupSpecifications(3 models)
| Model | Released | Parameters |
|---|---|---|
| LaMDA 137B | 2021-05 | 137B |
| LaMDA 8B | 2021-05 | 8B |
| LaMDA 2B | 2021-05 | 2B |
