PaLM 2
PaLM 2 launched at I/O in four sizes (Gecko, Otter, Bison, Unicorn), trained on 100+ languages and powering 25+ Google products.
Replaced PaLM with a more compute-efficient model with stronger multilingual, reasoning and coding ability. Became Bard's engine and the base for Med-PaLM 2 and Sec-PaLM. The technical report (arXiv 2305.10403, v1 2023-05-17) withholds detail on architecture and data.
- Date
- Wednesday, 10 May 2023
- Lab
- Google DeepMind
- Kind
- model
- Access
- closed API
Figures
| Measure | Value | Measured by |
|---|---|---|
| Model sizes | 4 (Gecko, Otter, Bison, Unicorn) Gecko small enough for mobile | company |
| Products powered | 25+ Bard, Workspace, Duet AI, Med-PaLM 2, Sec-PaLM | company |
Parameter count and training-token count not disclosed by Google.
Sources
This record was checked against its sources on 6 October 2026. How we check