Grok-1 (base)

Overview

xAI's first open-weight release: a 314B MoE with about 86B active parameters under Apache 2.0. Base model only — no official instruction tuning shipped.

When to pick this model

• Research into large-scale MoE architectures
• Custom fine-tuning projects with significant GPU budget
• Historical reference for xAI's open-weight lineage
• Apache-licensed base for downstream instruct training

VRAM requirements by quantization

Quantization	VRAM required
Q4_K_M (recommended)	188 GB
Q5_K_M	225 GB
Q8_0	335 GB
FP16 (no quantization)	630 GB

VRAM figures include model weights plus a typical 8k KV cache and ~600 MB runtime overhead (Ollama / llama.cpp baseline). Add headroom for higher context lengths.

Strengths

• First open-weight model from xAI
• Apache 2.0 with full commercial freedom
• Efficient MoE design with top-2 routing across 8 experts
• Useful base for community fine-tunes

Limitations

• Around 188 GB VRAM at Q4
• Raw base weights — no official instruct variant
• Comprehensively outpaced by Grok 2 and beyond
• Limited community fine-tunes vs. Llama or Qwen

Architecture & training

Architecture: MoE · 314B total / 86B active · 8 experts, 2 active · xAI

Training: xAI — first xAI open-source model. Raw weights released without official fine-tuning.

Quick start

# Non disponible via Ollama — poids HuggingFace uniquement

Or use the open-source MCP server to query this model from Claude Desktop, Cursor, or any MCP-compatible client.