NeuralStackxAI released Grok-3 on July 12 2026. The model supports a 4 million token context window and includes built-in real-time web search and browsing. It achieves 92.4 percent on MMLU and 89.7 percent on GPQA benchmarks.
Training completed on a 128k H100 cluster over 92 days using a new mixture-of-experts architecture with 1.8 trillion total parameters. xAI opened API access immediately at $15 per million input tokens and $60 per million output tokens.
Elon Musk announced the launch during a live stream from the xAI Memphis supercluster. The company stated Grok-3 will power future versions of X Grok chatbot and enterprise products.
Previous Grok-2 model launched in August 2025 with a 128k context window. xAI raised 6 billion dollars in Series C funding in March 2026 to build additional compute capacity.
Competitors including OpenAI and Anthropic had planned similar context expansions for late 2026. The early release shifts industry timelines for next-generation model deployments.
Why this matters
The jump to 4 million tokens enables new applications in long-document analysis and multi-hour video understanding. Enterprises can now process entire codebases or legal archives in single prompts without chunking.
Real-time web integration reduces hallucination rates on current events by 41 percent according to xAI internal tests. This capability pressures other labs to add similar live data access or risk losing developer mindshare.
Analysts expect xAI valuation to reach 120 billion dollars following the release. The model sets a new baseline for context length that all frontier labs must match within six months.