SpaceXAI Releases Grok 4.7: A Larger Base Model at the Same $2/$6 Price as Grok 4.6

SpaceXAI has released Grok 4.7, a new flagship model for coding, agentic work, and knowledge tasks. The company says it sits on a larger base model than Grok 4.6 and got a longer reinforcement learning run, but it keeps the same price and speed profile as the earlier version.

The headline for developers is pretty simple: you can call grok-4.7 now through the xAI API, Cursor, Grok Build, OpenRouter, Vercel, and Cloudflare. The docs list a 500,000-token context window, text and image input, text output, and four reasoning settings, with high as the default and xhigh at the top end. It also supports function calling, web search, X search, and code execution.

SpaceXAI says the model changed in four ways over Grok 4.6: a new larger base, a longer RL run on harder tasks, better self-verification and long-context handling, and native Grok Bot harness support. The company says the training mix leaned into problems that can take many hours to finish, which is a nice way of saying this was not just a quick polish pass.

On the benchmark sheet, Grok 4.7 looks strongest when the task feels like real work. SpaceXAI says it improves over Grok 4.6 in every row at xhigh effort, with the biggest jump on Terminal-Bench 4.0, where it moved from 20.3% to 38.0%. It also took the top score in EEBench at 64.0%, and scored 19.6% on Harvey’s legal agent benchmark.

But it doesn’t sweep the board. Fable 5.1 Max leads four of the seven benchmarks shown, including Terminal-Bench 4.0, and GPT-5.6 Sol Max still has the best DeepSWE v1.1 score. SpaceXAI’s pitch is really price-performance: Grok 4.7 stays at $2 per million input tokens and $6 per million output tokens, while the competition in the table is pricier.

Safety is part of the launch too. SpaceXAI says Grok 4.7 has a new safeguard stack and is the strongest model it has tested for refusals and jailbreak resistance. It also topped LatchBio’s biosafety benchmark at 62.4%, and on the company’s HackerBench v0.3, 3.3% of risky dual-use prompts got through. A US-only regional endpoint is available at a 10% premium, and Grok 4.7 Fast offers twice the output speed at twice the price, but only in Cursor and Grok Build.