Model

Grok-3 mini

CLOSED

SpaceXAI · Grok 3 family · released Feb 17, 2025

Cost-efficient reasoning model, ~90% cheaper than Grok 3; exposes a low/high reasoning_effort API parameter. Deprecated Feb 28, 2026 and again in the May 15, 2026 retirement wave.

ReasoningCodingVisionFunction callingTool useAgentic

1697.4

Elo · rank #107

Parameters

Undisclosed

Active params

Undisclosed

Context

131K tokens

Architecture

Mixture-of-Experts (size undisclosed; cost-efficient Grok 3 sibling)

License

Proprietary

Languages

—

API price (in/out)

$0.3 / $0.5

Modalities

text

Benchmark results

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

AIMEMath84.7%#69

best: GPT-5.2 · 100.0%

GPQA DiamondReasoning79.1%#93

best: GPT-6 Astra · 96.0%

Humanity's Last ExamReasoning11.1%#91

best: Claude Opus 5 · 64.7%

IFBenchReasoning45.9%#46

best: MiniMax M3 · 83.0%

LiveCodeBenchCoding69.6%#72

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

MATH-500Math99.2%#2

best: GPT-5 · 99.4%

MMLU-ProKnowledge82.8%#42

best: Claude Fable 5 · 91.5%

PutnamBenchMath0.0%#10

best: DeepSeek-Prover-V2 671B · 7.1%

SimpleQAKnowledge21.7%#23

best: GPT-4.5 · 62.5%

Run it locally

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

Input / M tok

$0.3

Output / M tok

$0.5

Grok 3 family

Elo progression across releases

Grok-3 miniFeb 20251697.4 Grok-3 (Standard)Feb 20251476.3 Grok-3 (Think)Feb 20251887.6

Resources

DOCSdocs.x.ai APIAvailable via SpaceXAI API

API price $0.3/$0.5 · each benchmark row carries its own source badge (see methodology)