SpaceX has released Grok 4.7, its latest large language model. The company integrated the model series and associated data centers following its acquisition of xAI. The new version utilizes a redesigned base model and an improved reinforcement learning workflow, which involved more complex training tasks.
In testing, Grok 4.7 was evaluated using CursorBench 4.0, achieving an average cost of $4.69 per task. This performance surpassed GPT-5.6 Sol and Fable 5.1. The model also outperformed Fable 5.1 on the Harvey Legal Agent Benchmark and EEBench, though it trailed behind GPT-6 Astra on the latter.
The model is compatible with the Grok Bot harness, allowing for the distribution of tasks among multiple AI agents for parallel processing and verification. Additionally, the system includes new safety features, setting performance records on the LatchBio and HackerBench benchmarks for blocking malicious requests.
Pricing for Grok 4.7 starts at $2 per million input tokens and $6 per million output tokens. SpaceX also offers a high-speed version for latency-sensitive workloads that doubles both processing speed and cost. This launch follows the release of the Grok Voice Transcribe 2.0 text-to-speech model.
