Event date · · xAI / Grok

Grok 3 Released: xAI Advances from Large-Scale Pretraining to Reasoning and Search Agents

FACT STATEMENT

xAI released Grok 3, Grok 3 mini, and their Think reasoning versions in February 2025, and previewed API and DeepSearch.

What happened

Grok enters the reasoning model competition, with products no longer emphasizing only real-time knowledge and expression style, but combining reinforcement learning, inference-time computation, long context, and search tools into agent capabilities.

Technical significance

Officially disclosed that Grok 3 was trained on the Colossus cluster, supporting 1M context; the Think version learns backtracking, verification, and multi-path solving through reinforcement learning, and can invest longer reasoning time per task.

Industry impact

Competition among frontier models now simultaneously tests pretraining scale, inference-time computation, and tool use, making single static benchmarks less able to explain real task capabilities.

Decision value

Buyers need to compare Grok 3's success rate, latency, and search evidence quality on real tasks, rather than directly adopting peak scores at release.

What to watch

Observe API stability, DeepSearch citation quality, tool call success rates, and cost curves under different reasoning budgets.

DECISION BRIEF

Turn the evidence into a decision.

See how AIGC.NEWS separates verified change, judgment, and the next signal to watch.