HAWTHORNE, September 24 — SpaceX introduced Grok 4.7 on Monday, September 21, calling it the company’s most capable large language model to date. The release lands less than a week after SpaceX’s previous model launch, a text-to-speech system it rolled out the Friday before. Grok began life at xAI Corp., the Elon Musk-founded startup.
Musk merged that company with xAI Inc. last year, and the combined organization was later folded into SpaceX, which picked up both the Grok model series and multiple artificial intelligence data centers in the process. Here’s what it means in plain terms: the rocket company now owns a frontier AI lab outright, data centers and all.
What the benchmarks actually show
SpaceX evaluated Grok 4.7 using CursorBench 4.0, a benchmark developed by Cursor, another of SpaceX’s recent acquisitions. Grok 4.7 completed the benchmark’s challenges at an average cost of $4.69 per task, ahead of GPT-5.6 Sol and Fable 5.1. SpaceX tested hardware-intensive versions of those two models, which prioritize output quality over cost-efficiency.
The company also ran Grok 4.7 on more widely used benchmarks. It handily outperformed Fable 5.1 on the Harvey Legal Agent Benchmark and EEBench, which contain legal and chip design tasks, respectively.
On EEBench, though, Grok 4.7’s score fell behind GPT-6 Astra, the latest LLM from OpenAI Group PBC. So the win column is real but not clean.
Why it got better
SpaceX credits Grok 4.7’s performance to a new base model. Think of it like a foundation poured before the framing goes up. LLM training runs comprise multiple phases, each improving a different aspect of the algorithm; a base model is the initial version of an LLM that emerges after the first phase, and AI providers often reuse base models across releases.
SpaceX says its engineers also enhanced Grok 4.7’s reinforcement learning workflow, an AI training method used to strengthen base models’ reasoning capabilities. Compared with its predecessor, Grok 4.7 received tougher training tasks and worked on them for longer.
The company built the model to be compatible with its Grok Bot harness, a collection of technical resources that let Grok 4.7 split complex work among multiple AI agents. Those agents can perform several different tasks in parallel, speeding up processing, and verify the accuracy of each other’s output. SpaceX also equipped Grok 4.7 with new safeguards, saying the model set records on LatchBio and HackerBench, benchmarks that test LLMs’ ability to block malicious biology research and cybersecurity requests, respectively.
What to watch
Grok 4.7 is available from $2 per million input tokens and $6 per million output tokens. For latency-sensitive workloads, SpaceX offers a version that processes prompts twice as fast and costs twice as much. The launch comes less than a week after SpaceX’s previous model release.
On Friday, the company introduced a text-to-speech model that offers double the accuracy of its predecessor and costs half as much. SpaceX says Grok Voice Transcribe 2.0 also outperforms several competitors’ entries in the text-to-speech category.
Two releases in under a week from a company best known for rockets is the thing to watch. The practical upshot for developers is a cheaper option on the price sheet and a new agent harness to build against. The open question is whether Grok 4.7’s edge holds on the benchmarks where it still trails.
