Grok 4 by xAI

Key takeaways

• Grok 4 is an xAI Transformer-based reasoning model with a 256k context window, 1.8 billion parameters, supporting text/image inputs, parallel tool calling, and structured outputs.
• The model demonstrates strong performance in easy-level science questions (AI2 Reasoning Challenge) with 99.04% accuracy, showcasing effective factual recall and understanding of common science concepts.
• Grok 4 struggles significantly with software engineering tasks, achieving only 7.67% accuracy on SWE-bench Lite, indicating limitations in applying abstract knowledge to practical scenarios.
• Known limitations include difficulty distinguishing subtle scientific concepts (e.g., confusing instruments for measuring length vs. time) and common failure modes like mixing up units of measurement.

Technical specifications

Context length

256k 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-4