xAI · 2025 · 02 · 17 · Model · ~1 min read
xAI launched Grok 3
What's actually new
- DeepSearch. A real-time web-search mode included free for the trial period — actually browses live, summarises with citations.
- Big think mode. Pauses to reason like OpenAI's o1 series. xAI's first thinking-style model.
- Trained on a much bigger pile of GPUs than Grok 2 — xAI's Memphis 'Colossus' cluster came online in late 2024.
If you want more
Worth knowing
- Launch tests were curated. On independent comparisons, Grok 3 sat between OpenAI o1-mini and full o1 — not above as the demo claimed.
- 'Most powerful' was a moment-in-time claim. OpenAI's o3 took the lead within two months.
Who should care
X Premium+ subscribers wondering what they're paying for. Developers comparing thinking models. Anyone curious about whether xAI's 'free speech' framing changes outputs.
What to do about it
Try DeepSearch through X Premium+ to see live web reasoning. Don't rely on Grok for sensitive topics yet — its safety filtering is more permissive than the others, which is a feature for some, a risk for others.
Honest take
Grok 3 caught xAI up to OpenAI on capability — not on track to lead. The DeepSearch idea was the more interesting bit: bringing live browsing into a reasoning loop, free, before the others had it. The bigger story was that xAI now has a real seat at the table with OpenAI, Anthropic, and Google. Whether that's good depends on how you feel about Musk's editorial choices flowing through it.
Sources
- xAI — Grok 3 announcementvendor
- Simon Willison — Grok 3 notesthird party
- Artificial Analysis — Grok 3 testingbenchmark
Last verified · 2026 · 05 · 05 · Found a fact wrong? corrections@aguidetocloud.com