Abstract:
xAI, Musk’s artificial intelligence company, officially released Grok 4.7, a new generation of large models, and positioned it as the most powerful programming and knowledge work model currently. The company stated that the new model not only improves the processing capabilities of complex tasks, but also maintains the same operating speed and price system as the previous generation products, and competes in the market at a lower cost.
According to information released by xAI, Grok 4.7 adopts a new larger-scale basic model and performs longer reinforcement learning optimization during the training process. Compared with previous versions, the new model is exposed to more complex training tasks, with a large number of tasks designed to last several hours to complete. This enables the model to achieve significant enhancements in long-term reasoning, self-verification, and complex task management.
xAI said that one of the biggest changes in Grok 4.7 is its stronger "self-test capability." Models can more proactively check their own output when answering questions, writing code, and performing complex knowledge tasks, thereby reducing error rates and improving the reliability of final results.
At the same time, the new model's ability to handle extremely long contexts has also been improved. When faced with large documents, multiple rounds of tasks, and long-duration project work, Grok 4.7 can maintain more stable information management capabilities without obvious performance degradation as the context grows like some traditional models.
In order to further improve the actual work experience, xAI also specially trains Grok 4.7 to natively understand the Grok Bot operating framework. This means that it is not only suitable for single-time question and answer scenarios, but also more suitable for participating in workflows as a long-running intelligent agent. The company believes this will improve model performance in general knowledge work, ongoing conversations, and enterprise-level tasks.
In addition to programming capabilities, Grok 4.7 is also endowed with stronger document and presentation creation capabilities. Officials stated that the model is now better able to generate professional reports, business documents and presentation materials, and hopes to further expand its application scope in the field of office automation.
In terms of performance, xAI announced a series of internal test results.
In the CursorBench 4.0 test, which is used to evaluate long-term software engineering capabilities, Grok 4.7 achieved a score of 46.3%, which was significantly improved compared to Grok 4.6's 40.4%. In the DeepSWE software engineering test, the new model achieved a score of 71%, continuing to maintain a competitive level close to the industry's leading products.
In the EEBench test in the field of electronic engineering, Grok 4.7 achieved a score of 64%, an increase of more than ten percentage points compared with the previous generation. For office tasks involving a large amount of professional knowledge and complex workflows, the model's performance in the AA Briefcase test also improved.
In terms of legal analysis, Grok 4.7 scored 19.6% in the Harvey Legal Agent Benchmark test, which is a further improvement compared to the previous generation product. In the field of medical reasoning, HealthBench Professional scores also improved from 48.5% to 56.7%.
In the field of general knowledge work, xAI particularly emphasized the two evaluation results of GDPval and AA Briefcase. The company believes that these tests are closer to the real work content of professionals such as lawyers, nurses, and financial analysts, and Grok 4.7 has reached the competitive level of current cutting-edge models in these scenarios.
Price is another important selling point of this release.
According to official data, the input price of Grok 4.7 remains at US$2 per million Tokens, and the output price is US$6 per million Tokens, consistent with Grok 4.6. Compared with some cutting-edge models that often cost several times or even ten times more, xAI emphasizes that Grok 4.7 can complete tasks of a similar level at a lower cost.
The company even describes it as a solution that "provides competitive performance at about half the price of models of the same level", hoping to further expand market share with the cost advantage.
In terms of security, xAI said that Grok 4.7 adopts a newly rebuilt security protection system. According to internal test results, the new model has reached the best level of the Grok series in terms of rejecting dangerous requests, resisting jailbreak attacks, and identifying high-risk content.
Particularly in sensitive areas such as cybersecurity and biosecurity, models are designed to help address legitimate defensive tasks while denying dangerous or malicious uses. Official data shows that Grok 4.7 achieved leading performance in some network security risk tests and biosecurity evaluations.
Currently, Grok 4.7 has been open to users through API and Grok platform. Developers can call this model for programming, knowledge Q&A, document writing, and intelligent agent application development.
With the official launch of Grok 4.7, xAI has further accelerated the pace of product iteration. From the release of Grok 4.5 in July this year, to the launch of Grok 4.6 in August, and now Grok 4.7, xAI has completed three major upgrades in less than three months. For the fiercely competitive artificial intelligence market, Grok 4.7 is not only a regular update, but also shows xAI’s strategic intention to challenge industry leaders through faster iterations, lower prices, and stronger work capabilities.
Comments