xAI shipped Grok 4 on July 9, 2025, introducing a flagship model with a 256K-token context window and an optional "Heavy" tier that coordinates five parallel agents for complex reasoning tasks. The update includes the specialized "Grok 4 Code" variant and integrates live data retrieval from X and other online sources.
- Grok 4 Heavy doubles the nearest rival on ARC-AGI-2 and clears a new bar on HLE, while matching graduate-level problem sets in math and formal logic.
- In coding benchmarks, it forms a top-tier trio with OpenAI's o3 and Claude Opus 4 on SWE-Bench, though it lags slightly on pass-at-1 Python metrics.
- Independent reviews highlight its strong STEM reasoning and real-time web integration, noting that the Heavy tier is expensive but unique in its multi-agent approach.
- Early adopters in finance, game development, and biomedical research are leveraging its large context window and structured output capabilities for rapid analysis and prototyping.
The combination of frontier reasoning scores, coding competence, and live data access positions Grok 4 as a significant competitor, particularly for users requiring up-to-date information and complex multi-step problem solving.