Model

PaLM (540B)

CLOSED

Google · PaLM family · released Apr 4, 2022

540B dense Pathways model; first to demonstrate strong chain-of-thought reasoning breakthroughs at scale.

ReasoningCodingVisionFunction callingTool useAgentic

380.5

Elo · rank #335

Parameters

540B

Active params

540B (dense)

Context

2K tokens

Architecture

Dense decoder-only Transformer (540B, Pathways)

License

Proprietary (Google; limited later API access, no public weights)

Languages

124+

API price (in/out)

No hosted API

Modalities

text

Benchmark results

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

ANLIReasoning48.4%#5

best: Phi-3-small (7B) · 58.1%

ARC-ChallengeReasoning53.0%#114

best: Llama 3.1 405B · 96.9%

ARC-EasyReasoning76.6%#33

best: Phi-3-medium (14B) · 97.7%

DROPReasoning69.4#46

best: Hunyuan-T1 · 93.1

GSM8KMath58.0%#118

best: Llama 3.1 405B · 96.8%

HellaSwagReasoning83.4%#57

best: Claude 3 Opus · 95.4%

LAMBADAReasoning77.9%#5

best: GPT-3 175B · 86.4%

MMLUKnowledge69.3%#150

best: OpenAI o3 · 92.9%

OpenBookQAReasoning53.4%#21

best: Claude 1 · 90.8%

PIQAReasoning82.3%#27

best: GPT-4o mini · 93.1%

RACE-HReasoning49.1%#12

best: Claude 3 Opus · 92.9%

StoryClozeReasoning84.6%#2

best: GPT-3 175B · 87.7%

SuperGLUEReasoning90.4%#2

best: UL2 20B · 90.7%

TriviaQAKnowledge81.4%#13

best: Sarvam-1 (2B) · 90.6%

WinoGrandeReasoning81.1%#32

best: PaLM 2 · 90.9%

Run it locally

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

Input / M tok

—

Output / M tok

—

PaLM family

Elo progression across releases

PaLM (540B)Apr 2022380.5

Resources

DOCSarxiv.org/abs/2204.02311

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