Model

Gopher

CLOSED

Google · Gopher family · released Dec 8, 2021

280B dense decoder-only Transformer; DeepMind's key data point in scaling research ahead of Chinchilla.

ReasoningCodingVisionFunction callingTool useAgentic

72.7

Elo · rank #392

Parameters

280B

Active params

280B (dense)

Context

2K tokens

Architecture

Dense decoder-only Transformer (280B)

License

Proprietary (Google DeepMind; research model, weights never released)

Languages

—

API price (in/out)

No hosted API

Modalities

text

Benchmark results

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

BIG-benchReasoning54.4%#2

best: Chinchilla · 65.1%

HellaSwagReasoning79.2%#87

best: Claude 3 Opus · 95.4%

LAMBADAReasoning74.5%#11

best: GPT-3 175B · 86.4%

MMLUKnowledge60.0%#206

best: OpenAI o3 · 92.9%

PIQAReasoning81.8%#30

best: GPT-4o mini · 93.1%

RACE-HReasoning71.6%#8

best: Claude 3 Opus · 92.9%

Social IQaReasoning50.6%#24

best: Apple DCLM-Baseline 7B · 82.9%

TriviaQAKnowledge63.6%#30

best: Sarvam-1 (2B) · 90.6%

TruthfulQAKnowledge29.5%#105

best: Phi-3.5-MoE (16x3.8B, 6.6B active) · 77.5%

WinoGrandeReasoning70.1%#102

best: PaLM 2 · 90.9%

Run it locally

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

Input / M tok

—

Output / M tok

—

Gopher family

Elo progression across releases

GopherDec 202172.7

Resources

DOCSarxiv.org/abs/2112.11446

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