xAI launches Grok 4.7 with a bigger base model at unchanged pricing

xAI released Grok 4.7 on September 21, 2026, a new flagship model for coding, agentic work and general knowledge tasks. The company says it runs on a larger base model than Grok 4.6, after a longer reinforcement learning run (training that rewards correct answers) weighted toward tasks that take many hours to finish. Despite the bigger base model, pricing and response speed stay the same as Grok 4.6.

The model is available now through the xAI API, and through Cursor, Grok Build, OpenRouter, Vercel and Cloudflare. It offers four reasoning-effort levels, low, medium, high and xhigh, with high as the default, and supports function calling, web search, X search and code execution. The API gives access to a 500,000-token context window, the amount of text the model can hold in view at once.

Benchmark gains, with limits

xAI’s launch benchmarks show Grok 4.7 ahead of Grok 4.6 on every measure the company reported. The biggest jump came on Terminal-Bench 4.0, a test of command-line task completion, where the score rose from 20.3% to 38.0%. EEBench climbed 11 points to 64.0%, the top score among the four models xAI compared. On Harvey’s legal agent benchmark, Grok 4.7 scored 19.6%, against 15.8% for Grok 4.6 and 6.7% for Fable 5.1 Max. On HealthBench Professional, a test of medical knowledge, Grok 4.7 scored 56.7%, trailing GPT-5.6 Sol Max at 60.5% and Fable 5.1 Max at 62.1%.

Grok 4.7 does not win everywhere. xAI’s own table puts Fable 5.1 Max ahead on four of seven benchmarks, including a 57.9% Terminal-Bench score, and GPT-5.6 Sol Max ahead on the DeepSWE v1.1 coding benchmark at 72.7%. xAI notes Grok 4.7’s DeepSWE figure was measured at a lower reasoning-effort setting than its other scores, so that comparison is not exact. On GDPval, a test of professional knowledge work, Grok 4.7 scored 1,695 Elo at its top reasoning setting, up from 1,605 for Grok 4.6, while Fable 5.1 Max led at 1,735. xAI says Fable 5.1 Max costs about five times more per input token and eight times more per output token than Grok 4.7; GPT-5.6 Sol Max costs roughly twice as much on input and more than three times as much on output.

xAI says Grok 4.7 ships with a new safeguard stack and calls it the strongest model it has tested for refusals and jailbreak resistance. The model topped a biosafety benchmark run by LatchBio at 62.4%. On HackerBench v0.3, xAI’s own test of risky cybersecurity prompts, 3.3% of dual-use requests got past its filters; the company says legitimate security research is rarely blocked. Select cybersecurity partners now get invite-only access to the model’s red-team capabilities for defensive research.

Grok 4.7 costs $2 per million input tokens and $6 per million output tokens, unchanged from Grok 4.6. It is the default model in Grok Build and available on all Cursor plans. A separate version, Grok 4.7 Fast, runs on faster infrastructure at twice the output speed and twice the price, but works only in Cursor and Grok Build. A US-only endpoint keeps inference inside the United States for a 10% premium. What remains untested is how the benchmark gains hold up outside xAI’s own figures, and how the new safeguard stack performs once outside researchers get to try it.