xAI has introduced Grok 4.7, a new flagship model aimed at software development and professional knowledge work. The company frames the release as a practical upgrade over Grok 4.6 rather than a wholesale change in pricing or serving speed.
According to the announcement, the model stays with demanding assignments for longer, reviews its own output more carefully, and ships with tighter safety / overall security calibration than earlier Grok versions.
The technical story behind the jump is relatively straightforward.
Grok 4.7 is built on a larger base model than its predecessor and received a longer reinforcement-learning run.
Training emphasized harder problems that can take many hours to finish, not just short prompts.
xAI says that mix improved self-checking and long-context handling.
The model was also trained to work natively with the Grok Bot harness, which the company expects to help with conversation, document drafting, and general office-style work.
Benchmark numbers published with the launch show gains on several coding and professional-work tests.
On CursorBench 4.0, which focuses on longer software-engineering tasks, Grok 4.7 scored 46.3 percent, compared with 40.4 percent for Grok 4.6.
On DeepSWE v1.1 it reached 71.0 percent at high effort, up from 65.2 percent.
Terminal-style work also improved: Terminal-Bench 4.0 rose from 20.3 percent to 38.0 percent on xAI’s reported harness.
Electrical-engineering scores on EEBench moved from 53.0 percent to 64.0 percent.
For multi-hour office tasks, AA Briefcase v1.1 climbed from 1,546 to 1,657.
Those results put the model near other frontier systems on some tests and behind them on others, depending on the evaluation and effort setting.
xAI is selling the model as a price-performance option. Standard API pricing starts at $2 per million input tokens and $6 per million output tokens, matching Grok 4.6.
The company says that combination is faster and cheaper than several comparable frontier models.
A faster serving option is available at twice the output speed and twice the price for latency-sensitive work.
Context remains large, with reports of a 500,000-token window and support for text and image input, tool use, search, and code execution.
Availability is immediate in several developer channels.
Grok 4.7 is live in Cursor and Grok Build, xAI’s coding-agent product.
It is also offered through the Grok API, third-party coding harnesses, model routers, and cloud platforms.
That distribution suggests the company wants the model used as an agent that can stay on a repository, a terminal session, or a document workflow rather than answering isolated questions.
Safety is another part of the pitch. xAI says Grok 4.7 uses a new safeguard stack and is the strongest Grok model so far on refusal quality and jailbreak resistance, while keeping refusals low for legitimate security and research work.
In dual-use areas such as cybersecurity and biology, the company reports stronger utility on benign tasks and stronger refusal on dangerous ones.
The release arrives only weeks after Grok 4.6 and continues xAI’s rapid iteration cycle.
For teams already using Grok in editors and agent stacks, the main question is whether the extra persistence and self-checking justify switching at the same token price. For everyone else, Grok 4.7 is another reminder that coding and long-horizon knowledge work have become the main arena where frontier labs now compete.