PaLM API / PaLM 2

Caution

Legacy. PaLM and PaLM 2 were superseded by Google’s Gemini models (Gemini replaced PaLM as Google’s flagship language-model line in 2023, per Wikipedia). Use the Gemini API or Vertex AI for new work; see Gemini 3 series. The exact PaLM API / text-bison / chat-bison retirement dates were not verified (see Open items).

PaLM (Pathways Language Model) is a family of large language models from Google (Alphabet). The original 540-billion-parameter PaLM was announced in April 2022 and stayed private until Google opened an API in March 2023. PaLM 2 was announced in May 2023 (Google I/O); it is reported (not officially confirmed) to be a 340-billion-parameter model trained on 3.6 trillion tokens (Wikipedia, secondary source).

Variants and surfaces

  • PaLM API / Vertex AI: developer access through the PaLM API (MakerSuite) and Google Cloud Vertex AI. Model IDs were text-bison (single-turn text) and chat-bison (multi-turn chat); aliases of this note keep those names so video titles match.
  • Med-PaLM / Med-PaLM 2: medical fine-tunes (per Wikipedia, reported to perform at expert level on U.S. medical licensing exam-style questions; not verified here).
  • PaLM-E (embodied, vision plus robotics) and AudioPaLM (speech-to-speech translation; announcement date not verified).
  • Sec-PaLM: security-analysis variant reportedly announced by Google Cloud in 2023 (not verified).

Why it still matters

  • Of historical interest for 2022-2023 LLM comparisons and for older tutorials that use text-bison/chat-bison.
  • Migrating old code: replace PaLM API calls with the Gemini Developer API or Vertex AI Gemini endpoints; Google’s current docs compare these two surfaces at https://ai.google.dev/gemini-api/docs/migrate-to-cloud.
  • Billing was per usage via Vertex AI; no prices recorded here (see the Google Cloud pricing pages).

Caveats (general, opinion)

Like any LLM, PaLM 2 could hallucinate and reflect training-data bias; outputs of domain variants such as Med-PaLM 2 would need human review in clinical settings.

Sources

Open items

  • Official retirement dates for the PaLM API and text-bison/chat-bison models not found (the Vertex AI model-versions page fetched did not list them).
  • PaLM 2 parameter count is a press/Wikipedia report, not a Google statement.
  • Sec-PaLM and Med-PaLM 2 details not verified against primary sources.