Key facts
- xAI released Grok 4.7 on Monday, after multiple delays since late July.
- The model scored below OpenAI's GPT-6 Astra and Anthropic's Claude Fable 5.1 on several benchmarks.
- Grok 4.7 has 2.1 trillion parameters and was trained with supplemental data from SpaceX engineering.
- xAI stated Grok 4.7 is faster and cheaper than Grok 4.6.
- The model is available via the Grok app, Cursor, Grok Build, and the xAI API.
Elon Musk's AI firm xAI released Grok 4.7 on Monday, its most advanced model to date, after a series of delays since late July. The company stated the new model is a "notable improvement over Grok 4.6 at the same price and speed," featuring enhanced safety guardrails and improved problem-solving capabilities.
Grok 4.7 boasts 2.1 trillion parameters, a 40% increase from its predecessor, and incorporates supplemental training data from SpaceX's engineering logs, aiming for better reasoning about physical systems. The model is now live across xAI's platforms, including the Grok app, Cursor, Grok Build, and the xAI API.
Despite xAI's claims of improvement, benchmark scores indicate Grok 4.7 trails behind competitors. On the GDPval benchmark, which measures performance on economically valuable knowledge work, Grok 4.7 scored 1695, falling short of Anthropic's Claude Fable 5.1, which achieved 1735. Similarly, on the AA-Briefcase test for multi-hour office tasks, Grok 4.7 scored 1657 against Fable 5.1's 1678. On EEBench, an electrical-engineering benchmark, Grok 4.7 placed second to OpenAI's GPT-6 Astra.
xAI has consistently positioned its models as a more affordable alternative to top-tier offerings from OpenAI and Anthropic, betting on a "good enough, cheap, and everywhere" strategy for broader adoption. Musk had previously suggested Grok 4.7 would be "roughly on par" with Anthropic's Claude Opus 5.0, with multimodal capabilities still under development.