Overview
- Tool name: PaLM 2 (Retired)
- Developer: Google
- Official website: https://ai.google/discover/palm2
- Category: AI Models
- Status: Retired across Google's product lines; legacy Vertex AI endpoints fully shut down on June 24, 2026
PaLM 2 was Google's flagship large language model announced at I/O in May 2023, built for the GPT-4 era. It shipped in four sizes - Gecko (on-device), Otter, Bison and Unicorn (frontier) - with multilingual pretraining across 100+ languages and markedly better reasoning than the original PaLM. It powered Bard (today's Gemini app), Duet AI for Workspace and MedPaLM 2, and was the backbone of Google AI before Gemini existed.
It has been fully superseded by the Gemini family. Google's consumer products (the Gemini app, Workspace AI, Search AI Overviews) all run on Gemini models. Vertex AI began transitioning in late 2024, the legacy Vertex AI SDK's generative module was deprecated in June 2025, and the final shutdown came on June 24, 2026. There is no reason to start any new project on PaLM 2.
Where PaLM 2 fell short (vs Gemini)
- An 8K-token context window (9K for Bison) versus Gemini Pro's million-token window - practically unusable for modern RAG
- Text-only: no vision, audio or tool-use capabilities
- No native prompt caching or newer throughput features; Gemini Flash tiers are several times faster
Migration notes
- Existing integrations: Move to Google's unified Gen AI SDK with Gemini Flash (speed/cost) or Gemini Pro (reasoning/long context) models
- Prompts: PaLM-era prompts likely need retuning on Gemini - watch for prompt drift
- Budget: Gemini Flash Lite runs as low as $0.10 per million input tokens, cheaper than maintaining legacy PaLM 2 endpoints with several times the throughput
Historical significance
PaLM 2 was the pillar of Google AI in the pre-Gemini era, and its size-tiering approach (Gecko through Unicorn) shaped the Flash/Pro product split Gemini uses today. For understanding Google's model lineage, it is a chapter you cannot skip.
Frequently asked questions
Is PaLM 2 still available?
No. Vertex AI PaLM 2 endpoints shut down on June 24, 2026, and all current Google products run on Gemini.
What is the relationship between PaLM 2 and Gemini?
Gemini is PaLM 2's successor. PaLM 2 powered Bard and Duet AI; from late 2023 Gemini progressively took over, and all post-merger Google DeepMind development targets Gemini exclusively.
Which model should I migrate to?
Everyday workloads: Gemini Flash tiers (best speed and cost). Reasoning and long documents: Gemini Pro (million-token context). Both via the unified Google Gen AI SDK.
*This page is not affiliated with Google. Retirement timeline summarized as of October 2026; confirm with Google Cloud's official documentation.*