Skip to main content
February 17 · 2025

Grok-3 — Flagship Leap on ~200k-GPU Colossus

Pushed frontier model performance with explicit focus on scientific discovery and maximum truth-seeking. 'Think mode' and DeepSearch bring AI closer to accelerating humanity's understanding of reality at machine scale.

Grok-3 (Feb 17, 2025): Flagship leap. Trained with ~10x compute of predecessor on Colossus (~200k GPUs at time). Outperformed or competed with leading models on math/science benchmarks (AIME, GPQA). 'Think mode' / reasoning (step-by-step like o-series). DeepSearch (web + X). Mini variant for speed. Daily improvements noted. API access. Grok-4 series (2025 onward): Grok 4 / 4 Heavy (July 2025) — further scaling, multi-agent. Grok 4.1 (Nov 2025) — reasoning, multimodal, reduced hallucinations, emotional intelligence. Grok 4.20 / 4.3 updates (late 2025–2026) with agentic/tool-calling, coding focus (Grok Build, Grok Code Fast). Fast variants for enterprise. Grok 5 teased at massive scale (trillions params roadmap).

Source: x.ai

Story
Post story on X

Read it in the full timeline

The story in depth

Independent educational fan project. Not affiliated with Space Exploration Technologies Corp. (SpaceX) or xAI Corp.