xAI launches Grok 3 with a Think button and Big Brain mode
At its 2025-02-17 livestream (post dated 2025-02-19) xAI launched Grok 3 with a "Think" button and "Big Brain" mode, saying the reasoning models were trained "using reinforcement learning at an…
- Date
- 17 February 2025
- Who
- xAI
- People
- Elon Musk, Igor Babuschkin
- Confidence
- Medium (company-reported; independent confirmation lagged)
- Deep dive
- Reasoning II, from o1 and o3 to DeepSeek-R1 and the labs that replicated them
Tier: Supporting · Significance: 3/5 · Org(s): xAI · People: Elon Musk, Igor Babuschkin · Confidence: Medium (company-reported; independent confirmation lagged) At its 2025-02-17 livestream (post dated 2025-02-19) xAI launched Grok 3 with a "Think" button and "Big Brain" mode, saying the reasoning models were trained "using reinforcement learning at an unprecedented scale" on the Colossus cluster, with 10x the compute of earlier frontier models; xAI reported AIME 2025 93.3% (at consensus-of-64) and GPQA 84.6% (xAI). A controversy followed when OpenAI staff pointed out xAI's chart omitted o3-mini-high's consensus@64 number and that at single-attempt (@1) Grok 3 trailed (TechCrunch, 2025-02-22). It arrived 158 days after o1-preview and 28 after R1. xAI's follow-up, Grok 4 (2025-07-09), claimed RL "at pretraining scale" (B08-32).