Model

Grok 4 Heavy

CLOSED

SpaceXAI · Grok 4 family · released Jul 9, 2025

Multi-agent 'Heavy' configuration of Grok 4, consumer-only via the $300/month SuperGrok Heavy tier; no public API model slug.

ReasoningCodingVisionFunction callingTool useAgentic

2253.2

Elo · rank #63

Parameters

Undisclosed

Active params

Undisclosed

Context

256K tokens

Architecture

Grok 4 base model served via parallel multi-agent test-time compute (reports describe 8-32 agents cross-checking answers)

License

Proprietary

Languages

—

API price (in/out)

No hosted API

Modalities

text · vision

Benchmark results

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

AIMEMath100.0%#3

best: GPT-5.2 · 100.0%

ARC-AGI-2Reasoning16.2%#39

best: GPT-6 Astra · 95.0%

FrontierMath (Tier 4)Math2.1%#13

best: GPT-5.6 Sol · 83.0%

GPQA DiamondReasoning88.9%#41

best: GPT-6 Astra · 96.0%

Humanity's Last ExamReasoning44.4%#21

best: Claude Opus 5.5 · 67.7%

LiveCodeBenchCoding79.4%#53

best: DeepSeek-V4-Pro (Think Max) · 93.5%

USAMOMath61.9%#4

best: Claude Opus 4.8 · 96.7%

Run it locally

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

Input / M tok

—

Output / M tok

—

Grok 4 family

Elo progression across releases

Grok-4Jul 20251895.2 Grok 4 HeavyJul 20252253.2

Resources

DOCSdocs.x.ai

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