Model

PaLM 2

CLOSED

Google · PaLM 2 family · released May 10, 2023

Powered Bard in 2023; basis for Med-PaLM 2 and Codey; PaLM API decommissioned Aug 15 2024.

ReasoningCodingVisionFunction callingTool useAgentic

792.2

Elo · rank #253

Parameters

Undisclosed

Active params

—

Context

8K tokens

Architecture

Dense Transformer (decoder, mixture-of-objectives pretraining)

License

Proprietary (Google Cloud Vertex AI)

Languages

—

API price (in/out)

No hosted API

Modalities

text

Benchmark results

Bar shows position within the tracked field; marker = field best

ARC-ChallengeReasoning95.1%#5

best: Llama 3.1 405B · 96.9%

BIG-Bench HardReasoning78.1%#38

best: ERNIE 4.5 300B-A47B · 94.3%

DROPReasoning85#15

best: Hunyuan-T1 · 93.1

GSM8KMath80.7%#80

best: Llama 3.1 405B · 96.8%

HellaSwagReasoning86.8%#22

best: Claude 3 Opus · 95.4%

MATH-500Math34.3%#155

best: GPT-5 · 99.4%

MGSMMath72.2%#33

best: OpenAI o4-mini · 93.7%

MMLUKnowledge78.3%#105

best: OpenAI o3 · 92.9%

OpenBookQAReasoning58.5%#16

best: Claude 1 · 90.8%

PIQAReasoning85.0%#10

best: GPT-4o mini · 93.1%

RACE-HReasoning62.3%#9

best: Claude 3 Opus · 92.9%

TriviaQAKnowledge86.1%#8

best: Sarvam-1 (2B) · 90.6%

WinoGrandeReasoning90.9%#1

best: this model · 90.9%

Run it locally

Closed weights — available via API only. No local deployment.

Input / M tok

—

Output / M tok

—

PaLM 2 family

Elo progression across releases

PaLM 2May 2023792.2

Resources

DOCSai.google/discover/palm2

API price — · each benchmark row carries its own source badge (see methodology)