Grok 4.7 Performance and Cost Analysis
Matthew Bermango watch the original →
the gist
Grok 4.7 is a highly cost-effective near-frontier model that excels in legal tasks and general knowledge work, though it currently trails top-tier models like Fable 5.1 and Astra Max in coding benchmarks and context window size.
Performance and Benchmarking
Grok 4.7 demonstrates strong performance across various domains, though it does not consistently lead the market. In coding-specific benchmarks like Cursorbench 4.0, Grok 4.7 achieves a 46.3% score at its highest thinking effort, trailing behind Fable 5.1 and Astra Max. While it underperforms on terminal-based coding tasks (38%) compared to Fable 5.1 (57.9%) and Astra Max (58.2%), it dominates in legal work with a 19.6% score, significantly outperforming competitors like GPT 5.6 Soul (2.5%) and Astra (5.4%). The model is currently ranked fifth on the Artificial Analysis intelligence index, placing it just behind Muse Spark 1.3.
Cost and Efficiency
The primary value proposition of Grok 4.7 is its aggressive pricing, which is driven by xAI's significant compute surplus and a strategic focus on cost-per-task efficiency. It is priced at $2 per million input tokens and $6 per million output tokens. This makes it roughly half the cost of GPT 5.6 Soul and significantly cheaper than Fable 5.1. While it is not the absolute frontier model, its high intelligence density makes it a viable choice for enterprise knowledge work where absolute peak performance is less critical than cost-effective automation. Notably, the model is limited to a 500K context window, whereas most competing frontier models offer 1M tokens.