xAI Releases Grok 4.7 With Focus on Coding and Long-Running Tasks

xAI has released Grok 4.7, its latest artificial intelligence model aimed at coding, agent-based tasks and professional knowledge work. The model became available on September 21, 2026, following a delayed rollout that had originally been expected earlier in the month.

xAI describes Grok 4.7 as its most capable model for coding and knowledge work. The company says the model uses a new, larger base model and underwent a longer reinforcement-learning process focused on more difficult tasks.

Grok 4.7 Targets Longer and More Complex Tasks

One of the main changes in Grok 4.7 is its training approach. xAI says the model was trained on a harder mixture of problems, including tasks that can require several hours to complete.

The company also says Grok 4.7 has been designed to spend more time checking its results and handling longer contexts. It was additionally trained to work with the Grok Bot environment, with the aim of improving its performance on conversational and general knowledge tasks alongside programming.

The model supports a 500,000-token context window, while developers can select different reasoning levels ranging from low to xHigh.

Pricing Remains at $2 and $6

Grok 4.7 retains the standard token pricing used for Grok 4.6.

Pricing	Grok 4.7
Input below 200K prompt tokens	$2 per million tokens
Cached input below 200K	$0.50 per million tokens
Output below 200K prompt tokens	$6 per million tokens
Input above 200K prompt tokens	$4 per million tokens
Cached input above 200K	$1 per million tokens
Output above 200K prompt tokens	$12 per million tokens

The model therefore keeps the same headline pricing as its predecessor for prompts below the 200,000-token threshold. xAI also offers a faster Grok 4.7 variant at twice the standard token rates through Cursor and Grok Build.

Because token consumption can vary significantly between tasks, the listed price alone does not determine the total cost of using an AI model.

Benchmark Results Show Gains Over Grok 4.6

xAI’s launch benchmarks show Grok 4.7 improving on Grok 4.6 across several tests.

On CursorBench 4.0, Grok 4.7 recorded 46.3%, compared with 40.4% for Grok 4.6. On Terminal-Bench 4.0, its score increased from 20.3% to 38.0%. The model also recorded 64.0% on EEBench and 19.6% on the Harvey Legal Agent Benchmark in xAI’s published comparison.

However, these results do not mean Grok 4.7 leads every competing model. For example, the same published comparison shows another model ahead of Grok 4.7 on Terminal-Bench 4.0 and several other tests. The benchmark figures should therefore be viewed as reported results under specific testing conditions, rather than a universal ranking of AI models.

Independent testing has also found that results can change depending on the tools and agent framework used with the model, highlighting the importance of evaluating models in real-world environments.

New Safety Measures

Grok 4.7 also introduces what xAI describes as a new safeguard system.

In xAI’s reported testing, the model achieved 62.4% on LatchBio’s biosafety benchmark. On HackerBench v0.3, a test focused on potentially risky cybersecurity tasks, xAI reported that 3.3% of risky dual-use prompts were allowed through.

xAI has also opened invite-only access to some cybersecurity partners so they can use Grok 4.7’s red-team capabilities for defensive security research.

These figures come from xAI’s own evaluations and should not be interpreted as universal measurements of the model’s safety across every possible use case.

Where Grok 4.7 Is Available

Developers can access Grok 4.7 through the xAI API, Cursor and Grok Build. The model is also available through model-routing and cloud platforms including OpenRouter, Vercel and Cloudflare.

Grok 4.7 Fast uses the same underlying model on faster infrastructure, although xAI lists it at twice the standard token rates. The fast version is currently available through Cursor and Grok Build rather than the public xAI API.

A Delayed Release

The launch followed several changes to the expected release schedule.

Elon Musk had previously indicated that Grok 4.7 would arrive earlier in September. Additional development work was subsequently required before the model was released on September 21.

The delay is consistent with xAI’s emphasis on longer reinforcement-learning runs and additional work on how the model handles difficult tasks and verifies its responses.

Looking Ahead for Grok

Grok 4.7 is arriving as AI developers increasingly focus on systems capable of completing longer sequences of work rather than responding to individual prompts.

For xAI, the latest release places greater emphasis on coding agents, professional tasks, tool use and extended workflows. The company’s benchmark results show progress over Grok 4.6, although independent testing will be important for determining how those improvements translate into everyday development and knowledge-work applications.

Conclusion

Grok 4.7 is xAI’s latest model for coding, agent-based work and professional knowledge tasks. It introduces a larger base model, longer reinforcement-learning training and additional safeguards while maintaining the previous generation’s standard pricing for smaller prompts.

The company’s published benchmarks show improvements over Grok 4.6, but the results do not establish Grok 4.7 as the overall leader across AI models. Its performance can also vary depending on the benchmark, task and surrounding tools.

With availability across the xAI API, Cursor, Grok Build and other platforms, developers can now evaluate the model directly as it begins to be used for longer and more complex workloads.

FAQs

When was Grok 4.7 released?
xAI released Grok 4.7 on September 21, 2026.

How much does Grok 4.7 cost?
For prompts below 200,000 tokens, the listed rates are $2 per million input tokens and $6 per million output tokens. Higher prompt lengths have higher rates.

What is Grok 4.7 designed for?
xAI positions it for coding, agent-based tasks and professional knowledge work.

What is the Grok 4.7 context window?
The model has a 500,000-token context window.

Does Grok 4.7 outperform every competing AI model?
No single benchmark establishes that. xAI’s published results show improvements over Grok 4.6 on several tests, while other models score higher on some benchmarks.