SpaceX acquired xAI for $1.25 trillion in February 2026. Then Grok 4.20 launched with 4 parallel AI agents per query. Grok 5 is training on 1.5GW. Here is everything you need to know.
On February 2 2026, SpaceX acquired xAI in what is described as the largest merger in history, valuing the combined entity at $1.25 trillion. The context matters: xAI was burning approximately $1 billion per month while SpaceX generates $8 billion in annual profits. The acquisition was financial necessity as much as strategic ambition. What changed for users is the infrastructure backing: Colossus 2, the training supercluster powering Grok 5's development, now runs at 1.5 gigawatts and is expanding. The compute trajectory is unlike anything else in the industry. Grok 4.20 launched February 17 2026 as the first major product release under the combined entity, introducing multi-agent architecture as the headline capability. Grok 5 is currently training and xAI targets Q2 2026 launch.
Grok 4.20's defining innovation is four specialized AI agents running in parallel on every complex query before synthesizing a unified response. The agents are not separate models: they are specialized heads on the same shared backbone. Grok coordinates, Harper handles research, Benjamin handles logic and code, Lucas handles divergent creative thinking. Cross-verification between agents reduced Grok 4.1's hallucination rate from 12% to 4.2%. Grok 4.20 pushes this further through systematic disagreement resolution before the final response is returned. In practice this means complex research questions get a higher-quality synthesized answer than single-model responses at the cost of slightly higher latency — typically 15 to 45 seconds for Heavy mode queries versus under 3 seconds for standard queries.
Grok's access to live X (Twitter) data is its only capability that competitors literally cannot replicate. Ask Grok about something that happened on X three hours ago and it knows. DeepSearch synthesizes information from multiple web sources and X simultaneously, producing a cited report in 2 to 5 minutes. No other major AI chatbot has this level of real-time social data access. For monitoring trends, fact-checking viral claims before they reach news sites, or understanding public sentiment on a product launch in real time, Grok is in a category of its own. Perplexity matches Grok on general web search accuracy and citation quality but cannot access live X conversations. This distinction matters most for marketing, PR, financial analysis and news monitoring workflows.
Grok 4 Heavy hit 100% on AIME 2025 and 88.4% on GPQA Diamond — performances that surpass Claude Opus 4.5 and GPT-4o on these benchmarks. However benchmark leadership shifts constantly between frontier models. Choose Grok 4.20 when you need real-time X data access, multi-agent synthesis on complex research, or the 2M context window at a competitive price point. Choose Claude Opus 4.6 for the highest coding quality on real tasks (80.8% SWE-bench), nuanced writing and agentic workflows where reliability is the priority. Choose ChatGPT GPT-5.4 for the broadest ecosystem, DALL-E integration and established enterprise tooling. SuperGrok at $30/month is more expensive than Claude Pro or ChatGPT Plus at $20/month — justify the premium only if X data access or Heavy mode is a genuine workflow need.