Grok-3 Mini
Compact Grok-3 variant for cost-effective conversations.
About Grok-3 Mini
Grok-3 Mini punches well above its price point for structured reasoning — at $0.30 per million input tokens, it scored 95.8% on AIME 2024 and 80.4% on LiveCodeBench, numbers that rival models costing several times more. Its 0.69-second time-to-first-token is unusually fast for a reasoning model, making it a practical choice when latency actually matters in production. The standout differentiator is native access to live X data, giving it a real edge for tracking breaking news or social sentiment that static-knowledge models simply cannot match. Users consistently praise its concise reasoning traces and cost efficiency for math, logic, and quantitative work. That said, real-world coding reliability is a known weak spot — community reports diverge from the benchmark numbers, with users finding it unreliable on practical programming tasks outside controlled evaluations. If your workload centers on structured problem-solving, agentic pipelines, or monitoring fast-moving topics, Grok-3 Mini delivers serious capability at a price that keeps inference costs manageable. General-purpose coding or broad factual recall is better served elsewhere.
Best for
- Math, logic, and quantitative reasoning where benchmark accuracy (95.8% AIME) translates to real gains
- Real-time social sentiment and trending topic research via native X/Twitter data integration
- Cost-sensitive agentic pipelines and automation workflows that need function calling and structured outputs
- Long-context document analysis using its 128K token window
- Latency-sensitive reasoning tasks where a 0.69-second time-to-first-token matters
Specs & capabilities
How Grok-3 Mini stacks up — intelligence, speed, context, and modalities.
Intelligence
Medium
Speed
Medium
Context window
131,072 tokens
Knowledge cutoff
February 28, 2025
Frequently asked questions
How much does Grok-3 Mini cost?
Input tokens are $0.30 per million and output tokens are $0.50 per million — significantly cheaper than full Grok-3 and competitive with other small reasoning models.
What is the context window?
131,072 tokens (128K), which is enough to process large documents or multiple files in a single call.
What is Grok-3 Mini genuinely best at?
Structured math and reasoning tasks, quantitative analysis, and anything benefiting from real-time X data. Its AIME 2024 score of 95.8% is a standout for a model at this price tier.
Where does it fall short?
Real-world coding reliability is inconsistent — community feedback suggests it underperforms on practical programming problems despite solid benchmark numbers. Hallucination rates on topics outside X's trending sphere are also higher than Claude or ChatGPT.
How does it compare to full Grok-3?
Grok-3 Mini trades some raw capability for much lower cost and faster inference. It is the right choice when budget and latency matter more than squeezing out maximum performance.
Is there a reasoning mode, and does it change behavior?
Yes. The Reasoning variant produces longer chains of thought and can be significantly slower, generating far more output tokens than the standard mode. Enable it when accuracy on hard problems matters; skip it when speed is the priority.