Grok 4.7 Explained: Features, Performance, Capabilities & What’s New

Grok 4.7 is xAI’s latest AI model, introduced on September 21, 2026, with a stronger focus on coding, long-running tasks, professional knowledge work, self-verification, and agentic workflows.

Compared with Grok 4.6, the new model uses a larger base model and a longer reinforcement-learning training run focused on difficult tasks that can take hours to complete. xAI says Grok 4.7 is better at checking its own work, managing longer context, working with the Grok Bot environment, and handling general knowledge work.

For developers, the model starts at $2 per million input tokens and $6 per million output tokens. xAI also offers a faster variant at twice the output speed for twice the price. Grok 4.7 is available through the Grok API, Grok Build, Cursor, third-party coding harnesses, model routers, and cloud platforms.

Quick Summary

• Grok 4.7 launched on September 21, 2026, focusing on coding, reasoning, long-running tasks, and professional knowledge work.
• It introduces stronger self-verification, long-context management, and agentic capabilities compared with Grok 4.6.
• xAI reports improved results across coding, terminal, legal, clinical, office, and engineering benchmarks.
• API pricing starts at $2 per million input tokens and $6 per million output tokens.
• Grok 4.7 is available through the Grok API, Grok Build, Cursor, coding harnesses, model routers, and cloud platforms.
• Its strongest use cases include software development, AI agents, research, automation, and professional workflows.
• As of September 21, xAI’s consumer pricing page still lists Grok 4.6 with SuperGrok, so Grok 4.7 consumer access should not be assumed.

What Is Grok 4.7?

Grok 4.7 is xAI’s newest model designed primarily for coding and knowledge work. Rather than simply increasing general chatbot capabilities, the release emphasizes tasks where an AI system needs to reason over a problem for a longer period, use tools, produce professional work, and verify the result.

The model was trained using a larger base model than Grok 4.6 and a longer reinforcement-learning process built around harder, longer-duration tasks. xAI says the training mix was weighted toward problems that can take many hours to solve.

This makes Grok 4.7 particularly relevant to:

• Software engineering
• Agentic coding
• Long-running technical tasks
• Professional knowledge work
• Document and presentation creation
• Terminal-based workflows
• Research and analysis
• Legal and clinical reasoning tasks
• Electrical engineering
• AI agents using computer environments

Grok 4.7 also has native understanding of the Grok Bot harness, which xAI says improves its performance for conversational tasks and general knowledge work.

When Was Grok 4.7 Released?

xAI introduced Grok 4.7 on September 21, 2026.

The launch comes only a few weeks after the broader Grok ecosystem expanded around Grok Bot, Grok Build, connectors, coding tools, and other agentic capabilities.

The release follows several recent Grok developments:

Grok 4.6 remains important because it provides the immediate baseline for measuring the improvements in the new model. xAI’s launch comparison shows Grok 4.7 outperforming Grok 4.6 across several of its reported evaluations.

What Are the New Grok 4.7 Features?

The most important Grok 4.7 features are centered on long-horizon reasoning, verification, coding, agentic work and professional tasks.

1. Longer-Horizon Task Performance

One of the biggest changes is the model’s focus on tasks that do not have an immediate answer.

Instead of producing an answer after a short reasoning process, Grok 4.7 is designed to work through more complicated tasks for longer periods.

This matters for software engineering and professional workflows where the AI may need to:

1. Understand a large problem.
2. Break it into smaller tasks.
3. Work through multiple steps.
4. Use tools or a computer environment.
5. Check intermediate results.
6. Correct mistakes.
7. Produce a final deliverable.

This approach is closer to an AI worker or agent than a traditional question-and-answer chatbot.

2. Better Self-Verification

xAI specifically highlights improved ability to verify its own work.

Self-verification is important for coding and technical work because producing an answer is only one part of the task.

For example, a coding agent may need to:

• Write code
• Run tests
• Inspect failures
• Modify the implementation
• Run the tests again
• Check the final output

Grok 4.7 was trained with greater emphasis on these longer-running task patterns.

3. Improved Coding

Coding is one of the central use cases for Grok 4.7.

xAI’s launch data reports improvements over Grok 4.6 on both CursorBench 4.0 and DeepSWE v1.1.

Grok 4.7 achieved:

• 46.3% on CursorBench 4.0
• 71.0% on DeepSWE v1.1 at high effort

For comparison, Grok 4.6 scored 40.4% and 65.2%, respectively, in xAI’s published comparison.

These numbers are vendor-published benchmark results, so they should be interpreted as measurements from xAI’s selected evaluation setup rather than a universal ranking of coding ability.

4. Agentic Workflows

Grok 4.7 is designed to work particularly well in environments where the model can use tools and continue working through a task.

This connects directly with xAI’s broader agent ecosystem, including Grok Bot and Grok Build.

Grok Bot gives AI agents their own computer environment and allows them to work across tools and applications. xAI describes the agents as capable of handling jobs end-to-end and returning when human approval is required.

Grok 4.7’s native understanding of the Grok Bot harness is therefore significant. The model is not being developed in isolation from the agent environment in which it can operate.

5. Professional Knowledge Work

Grok 4.7 is not limited to programming.

xAI also reports improvements in tasks involving professional knowledge work, including document and presentation creation.

The company evaluated the model on tasks involving areas such as:

• Legal work
• Clinical reasoning
• Financial and office tasks
• Electrical engineering
• General professional knowledge work

On the AA Briefcase v1.1 evaluation for multi-hour office work, Grok 4.7 scored 1,657 compared with 1,546 for Grok 4.6 in xAI’s published results.

6. Longer Context Management

xAI says Grok 4.7 is better at managing longer context.

This is particularly useful when an AI system has to work with:

• Large codebases
• Long documents
• Research materials
• Extended conversations
• Multi-step agent histories
• Large project specifications

However, xAI’s launch announcement does not state a new numerical context-window size for Grok 4.7. Therefore, claims that Grok 4.7 has a specific 1-million- or 2-million-token context window should not be treated as confirmed unless xAI’s current model documentation specifies it.

That distinction is important because previous Grok models have had different context limits.

Grok 4.7 Performance: How Good Is It?

Grok 4.7’s performance is best understood through the tasks xAI targeted rather than through one overall benchmark score.

The company published results across coding, terminal work, office work, legal work, clinical reasoning and electrical engineering.

Grok 4.7 vs Grok 4.6

The Grok 4.7 DeepSWE result shown by xAI was measured at high effort.

The largest improvement in this table is on Terminal-Bench 4.0, where Grok 4.7’s reported score is substantially higher than Grok 4.6’s.

That result is particularly relevant for developers because terminal-based tasks require an AI model to work through multiple steps rather than simply generate code in a single response.

Grok 4.7 vs Other AI Models

xAI also compared Grok 4.7 with several other frontier models.

For example, its published results show:

Model	Input / 1M tokens	Output / 1M tokens	CursorBench 4.0
Grok 4.7	$2	$6	46.3%
Grok 4.6	$2	$6	40.4%
GPT-5.6 Sol	$4	$20	41.7%
Fable 5.1	$10	$50	51.8%

These figures come from xAI’s September 21 launch comparison and should be treated as provider-published results, not as an independent industry ranking. Different models can perform differently depending on prompting, reasoning effort, tools, harnesses, evaluation methodology and task distribution.

The pricing comparison is nevertheless notable: Grok 4.7’s published API token price is substantially below the two other models included in xAI’s table.

Grok 4.7 Pricing

Grok 4.7 API pricing starts at:

• $2 per million input tokens
• $6 per million output tokens

xAI also offers a fast variant that provides twice the output speed at twice the price.

That means developers should distinguish between the standard Grok 4.7 price and the faster variant when estimating API costs.

Is Grok 4.7 Included With SuperGrok?

This is an important point because the model launch and consumer subscription lineup are not currently identical.

As of September 21, 2026, xAI’s public consumer pricing page lists Grok 4.6 under SuperGrok rather than Grok 4.7. SuperGrok is currently listed at $30 per month, while SuperGrok Plus is listed at $100 per month.

Therefore, it would be inaccurate to automatically state that buying SuperGrok gives you Grok 4.7 access.

The current confirmed Grok 4.7 availability announced by xAI includes:

• Grok API
• Grok Build
• Cursor
• Third-party coding harnesses
• Model routers
• Cloud platforms

xAI also provides free access to Grok Build for trying the model in that environment.

Consumer availability can change, so users should check the live Grok model picker and xAI pricing page before subscribing specifically to access Grok 4.7.

Grok 4.7 API Pricing vs Grok 4.6

The launch announcement states that Grok 4.7 is served at the same price as Grok 4.6.

The published comparison lists:

Model	Input	Output
Grok 4.7	$2/M	$6/M
Grok 4.6	$2/M	$6/M

For developers, this means the primary change is improved model capability rather than a higher standard token price.

Developers should still consult the live API documentation because xAI can change model availability, context pricing, caching rates, or model-specific billing rules over time.

How Does Grok 4.7 Compare With Grok 4.6?

The easiest way to understand the difference is to think of Grok 4.6 as a strong general model for long-running agents and visual work, while Grok 4.7 pushes further toward extended coding and professional workflows.

The comparison does not mean Grok 4.6 immediately becomes irrelevant. Model selection depends on the application, available tools, latency requirements, context needs and access route.

What Can You Use Grok 4.7 For?

Software Development

Grok 4.7 is particularly suited to software engineering workflows.

Possible applications include:

• Debugging
• Code generation
• Refactoring
• Test creation
• Repository analysis
• Terminal tasks
• Code review
• Software architecture
• Long-running coding agents

Its improvements on CursorBench, DeepSWE and Terminal-Bench are especially relevant to this category.

Research and Knowledge Work

Grok 4.7 can also be used for longer professional tasks involving research, analysis and document creation.

A practical workflow might look like:

Research → Analyze → Draft → Verify → Revise → Deliver

This is different from using an AI model simply to generate a paragraph or answer a short question.

Documents and Presentations

xAI specifically reports improvements in document and presentation creation.

That could make the model useful for workflows involving:

• Reports
• Business documents
• Presentations
• Research summaries
• Structured analysis
• Professional communications

The actual quality of the final output still depends on the prompt, source material, tools and verification process.

Agentic Automation

Grok 4.7’s connection with Grok Bot makes it relevant to automated workflows where an AI system needs to operate inside a computer environment.

Grok Bot can work with applications, inboxes and other tools and has its own computer environment.

This creates a workflow closer to:

Instruction → Planning → Computer Interaction → Tool Use → Verification → Completion

rather than:

Prompt → Text Response

Cybersecurity Research

xAI also introduced a new safeguard stack with Grok 4.7.

The company reports that the model allows only 3.3% of risky dual-use prompts through on its HackerBench v0.3 evaluation and says it provides low refusal rates for legitimate security work. It has also begun giving selected cybersecurity partners invite-only access to red-team capabilities.

These are xAI’s own evaluation results, so they should be treated as vendor-reported evidence rather than an independent certification of cybersecurity safety.

What Are the Limitations of Grok 4.7?

Grok 4.7’s improvements do not remove the normal limitations of frontier AI systems.

Benchmark results are not universal

A benchmark score measures performance on a particular evaluation.

Real-world results can vary depending on:

• Prompt design
• Tools
• Agent harness
• Context
• Task complexity
• Reasoning settings
• Software environment
• Evaluation methodology

Consumer availability is still evolving

The model is clearly available through the API and several developer environments, but the current consumer pricing page still identifies Grok 4.6 as the model associated with SuperGrok.

This means users should not assume that a consumer subscription automatically provides Grok 4.7.

Long-running agents still require supervision

An AI agent that can work for longer periods can also make mistakes for longer periods.

For production systems, developers should use:

• Sandboxing
• Permission controls
• Human approval for sensitive actions
• Logging
• Automated tests
• Output validation
• Spending limits

API costs depend on usage

The headline $2/$6 pricing is per million tokens.

Actual costs depend on:

• Input volume
• Output volume
• Context length
• Number of agent steps
• Tool calls
• Repeated requests
• Fast-mode usage
• Application architecture

A long-running coding agent can therefore cost considerably more than a simple chatbot interaction.

Is Grok 4.7 Better Than Grok 4.6?

Grok 4.7 represents a documented improvement over Grok 4.6 on the benchmarks xAI published with the launch, including coding, terminal work, office tasks, legal work, clinical reasoning and electrical engineering.

However, “better” depends on the workload. The most meaningful change is that Grok 4.7 was explicitly trained around longer and harder tasks, with greater emphasis on self-verification and extended context management.

For developers building coding or agentic workflows, those changes are more relevant than a generic benchmark comparison.

Is Grok 4.7 Free?

Grok 4.7 is available through Grok Build for free, according to xAI’s launch announcement.

The API is usage-based, starting at $2 per million input tokens and $6 per million output tokens.

However, free availability in Grok Build should not be confused with unlimited access to the model across every Grok product.

xAI’s consumer Grok service has separate subscription plans and usage limits. The current pricing page lists Free, SuperGrok, SuperGrok Plus and other plans, with SuperGrok starting at $30 per month.

Who Should Use Grok 4.7?

Grok 4.7 is most relevant for users who need AI to perform longer, multi-step tasks rather than only answer short questions.

It is particularly relevant to:

• Software developers
• Coding-agent users
• Technical researchers
• AI application developers
• Professional knowledge workers
• Businesses experimenting with AI agents
• Teams building computer-use workflows
• Developers who need relatively low API token pricing

For ordinary users who mainly want chat, image generation, voice or general-purpose assistance, the consumer Grok product lineup remains a separate consideration

Conclusion

Grok 4.7 is a significant shift toward longer-running AI work rather than simply another incremental chatbot release.

Its biggest changes are the larger base model, longer reinforcement-learning training, stronger self-verification, improved long-context management and deeper integration with agentic environments such as Grok Bot.

For developers, the $2 per million input and $6 per million output token starting price makes the model particularly interesting for coding and agent workloads. At the same time, the distinction between API/developer availability and consumer subscription access is important: xAI’s current consumer pricing page continues to list Grok 4.6 for SuperGrok.

As Grok’s agent ecosystem expands, the more important question is increasingly not just how well Grok 4.7 answers a prompt, but how reliably it can work through a complex task, use tools, verify its output and complete the job.

Frequently Asked Questions

1. What is Grok 4.7?

Grok 4.7 is xAI’s September 2026 AI model focused on coding, long-running tasks, professional knowledge work, self-verification and agentic workflows. It uses a larger base model and longer reinforcement-learning training than Grok 4.6.

2. When was Grok 4.7 released?

Grok 4.7 was officially introduced by xAI on September 21, 2026.

3. How much does Grok 4.7 cost?

Grok 4.7 API pricing starts at $2 per million input tokens and $6 per million output tokens. xAI also offers a faster variant at twice the output speed for twice the price.

4. Is Grok 4.7 available in SuperGrok?

As of September 21, 2026, xAI’s consumer pricing page lists Grok 4.6 under SuperGrok rather than Grok 4.7. Grok 4.7’s confirmed launch availability includes the API, Grok Build, Cursor, third-party coding harnesses, model routers and cloud platforms.

5. Is Grok 4.7 good for coding?

Coding is one of Grok 4.7’s primary target areas. xAI reports improvements over Grok 4.6 on CursorBench 4.0, DeepSWE v1.1 and Terminal-Bench 4.0, although benchmark results should be interpreted within their specific evaluation setups.

6. Does Grok 4.7 have a 1-million-token context window?

xAI’s September 21 launch announcement says Grok 4.7 is better at managing longer context but does not publish a new numerical context-window limit in that announcement. A specific 1-million-token figure should therefore not be treated as confirmed for Grok 4.7 without corresponding current model documentation.

Also Read –

SuperGrok Pricing and Plans: What Features Do You Get?

Grok AI Features: Explore Everything Grok Can Do

Grok 4.6 Complete Guide: Capabilities, Pricing, Benchmarks, and Availability