xAI Unveils Grok 4.7 with Enhanced Performance for Coding and Knowledge Tasks
xAI has officially released Grok 4.7, its latest AI model designed to enhance coding and knowledge work capabilities. This new version boasts improved performance on long-running tasks and stronger safeguards compared to its predecessor, Grok 4.6.
The updated model utilizes a larger base model and has undergone an extended reinforcement learning phase, specifically targeting more complex tasks that can take hours to complete. As a result, Grok 4.7 has demonstrated significant improvements across various benchmarks published by xAI.
- CursorBench 4.0: Grok 4.7 scored 46.3%, up from Grok 4.6's 40.4%.
- Terminal Bench 4.0: The score increased to 38% from 20.3%.
- DeepSWE 1.1: Achieved a score of 71% at high reasoning effort.
- AA Briefcase: Scored 1,657, marking a significant improvement.
- EEBench: Achieved a score of 64%.
While Grok 4.7 outperformed Grok 4.6 in several areas, it still trails behind models like GPT 5.6 Sol and Fable 5.1 in some evaluations. The pricing for Grok 4.7 remains unchanged at $2 per million input tokens and $6 per million output tokens, with a faster version available for double the price.
Additionally, xAI has introduced a new safeguard stack for Grok 4.7, allowing only 3.3% of risky dual-use prompts through on HackerBench 0.3. The model also scored 62.4% on a biosafety benchmark from LatchBio. Selected cybersecurity partners will have access to the model’s red team capabilities for defensive research.
Grok 4.7 is accessible via Cursor, Grok Build, the Grok API, and various third-party coding tools and cloud platforms.
FAQ
What are the key improvements in Grok 4.7 compared to Grok 4.6?
Grok 4.7 features enhanced performance on long-running tasks, stronger safeguards, and improved scores across various benchmarks. It utilizes a larger base model and has undergone extended reinforcement learning, resulting in significant improvements in coding and knowledge tasks.
How does Grok 4.7 perform on benchmarks?
Grok 4.7 achieved notable scores on several benchmarks: 46.3% on CursorBench 4.0, 38% on Terminal Bench 4.0, 71% on DeepSWE 1.1 at high reasoning effort, 1,657 on AA Briefcase, and 64% on EEBench.
What is the pricing structure for Grok 4.7?
The pricing for Grok 4.7 remains unchanged at $2 per million input tokens and $6 per million output tokens. A faster version is available for double the price.
What safeguards have been introduced in Grok 4.7?
Grok 4.7 includes a new safeguard stack that allows only 3.3% of risky dual-use prompts through on HackerBench 0.3. Additionally, it scored 62.4% on a biosafety benchmark from LatchBio.
Where can users access Grok 4.7?
Grok 4.7 is accessible via Cursor, Grok Build, the Grok API, and various third-party coding tools and cloud platforms.
Comments
Comments are moderated before publish.
No comments yet — be the first.