SpaceXAI has released Grok 4.7, a new artificial intelligence model built for coding and knowledge work. The company calls it its most capable model to date.
The new model costs the same as Grok 4.6 and runs at the same speed. SpaceXAI says it is twice as fast as comparable models at half the price.
According to the company, Grok 4.7 works longer on hard tasks and checks its own work more carefully. It also comes with new safety tools.
What Changed in Grok 4.7
Grok 4.7 uses a new, larger base model than Grok 4.6. It was trained with a longer reinforcement learning run on harder tasks, many of which take hours to finish.
The company says the model is better at managing long context. It was also trained to work natively with the Grok Bot system, which helps with conversation and general knowledge work.
On CursorBench 4.0, a test focused on long coding tasks, Grok 4.7 scored 46.3%. That is up from 40.4% for Grok 4.6 and ahead of GPT-5.6 Sol at 41.7%. Fable 5.1 scored higher at 51.8%.
The model posted mixed results on other tests. It scored 64.0% on EEBench, an electrical engineering test, beating every model listed. On Terminal-Bench 4.0, it scored 37.6%, behind Fable 5.1 at 57.9%.
In legal work, Grok 4.7 scored 19.6% on the Harvey Legal Agent Benchmark. That was higher than GPT-5.6 Sol at 2.5% and Fable 5.1 at 6.7%.
On HealthBench Professional, which tests clinical reasoning, it scored 56.7%. Both GPT-5.6 Sol and Fable 5.1 scored higher.
SpaceXAI also said Grok 4.7 is better at making documents and presentations. On GDPval, it earned an Elo score of 1,695, compared with 1,735 for Fable 5.1 and 1,605 for Grok 4.6.
Safety and Cybersecurity
Grok 4.7 uses an entirely new set of safeguards. SpaceXAI says it is the strongest model it has tested at refusing harmful requests and resisting jailbreaks.
The model scored 62.4% on LatchBio's biosafety benchmark, which the company says is the top result. On HackerBench v0.3, SpaceXAI's own test for risky cyber tasks, it let through only 3.3% of risky prompts.
The company says the model rarely blocks legitimate security work. Select cybersecurity partners are getting invite-only access to its red-team tools for defense research.
Grok 4.7 is available now in Cursor and Grok Build. It can also be used through the Grok API, third-party coding tools, model routers, and cloud platforms.
Pricing starts at $2 per million input tokens and $6 per million output tokens. By comparison, GPT-5.6 Sol costs $4 and $20, while Fable 5.1 costs $10 and $50.
SpaceXAI also offers a faster version of the model. It delivers twice the output speed at twice the price.
