Grok 3 by xAI

Key takeaways

• Grok 3 is a Transformer-based text-to-text model from xAI, primarily designed for enterprise use cases such as data extraction, coding, and text summarization, with specialized domain knowledge in finance, healthcare, law, and science.
• The model demonstrates strong capabilities in elementary science reasoning, achieving 99.07% accuracy on the AI2 Reasoning Challenge - Easy dataset, indicating efficiency in extracting correct answers.
• Grok 3 struggles significantly with complex multi-step reasoning, quantitative questions, detailed biological/medical scenarios, adhering to specific output formats, and processing prompts with multiple constraints.
• The model scored a low 4.7% accuracy on 'Humanity's Last Exam', highlighting limitations in applying domain knowledge, performing accurate quantitative calculations, and processing complex, multi-constrained prompts.
• While Grok 3 provides verbose explanations, these often lack accuracy and can suffer from overconfidence or hallucination, especially in complex reasoning, even when the final answer is incorrect.

Technical specifications

Context length

131k tokens

Max output

—

Architecture

Transformer

Modalities

text->text

Parameters

~1800B

Open weights

No

Licence

MIT

Regions

USA

Vendor

xAI

Provider model key

x-ai/grok-3