What is Grok 4.7?
Grok 4.7 is xAI’s proprietary model for coding, knowledge work and longer-running tasks. Released September 21, 2026, it uses a larger base model than Grok 4.6 and additional reinforcement learning focused on problems that take many hours to complete.
xAI serves Grok 4.7 through its API, Grok Build, Cursor, third-party coding harnesses and supported cloud or model-routing platforms. The standard route starts at $2 per million input tokens and $6 per million output tokens for prompts below 200,000 tokens.
QUICK VERDICT: Grok 4.7 is xAI’s strongest general model for coding and knowledge work. It improves over Grok 4.6 without changing the standard short-context token prices, but requests at or above 200,000 prompt tokens cost twice as much, and its company benchmark results still need validation on real workloads.
Field | Verified value |
|---|---|
Provider | xAI |
Release date | September 21, 2026 |
License | Proprietary |
API model ID | grok-4.7 |
Context window | 500,000 tokens |
Modalities | Text model; current-events access requires search tools |
Knowledge cutoff | May 2026 |
Reasoning | Encrypted reasoning content returned through the Responses API |
Base pricing below 200K prompt tokens | $2 input, $0.50 cached input and $6 output per million tokens |
Long-context pricing | $4 input, $1 cached input and $12 output per million tokens |
Best fit | Coding, knowledge work and long-running tasks |
Last verified | October 2, 2026 |
What changed from Grok 4.6
Grok 4.7 changes the underlying model rather than merely tuning the Grok 4.6 serving profile. xAI says it uses a larger base model, a longer reinforcement-learning run, a harder task mixture, stronger self-verification and native understanding of the Grok Bot harness.
Area | Grok 4.7 change |
|---|---|
Base model | Larger underlying model than Grok 4.6 |
Training | Longer reinforcement-learning run on harder, multi-hour tasks |
Verification | Improved checking of intermediate work and final results |
Coding | Higher reported scores on CursorBench, DeepSWE and Terminal-Bench |
Knowledge work | Improved document, presentation, legal and clinical workflows |
Safeguards | New safeguard stack with stronger refusal and jailbreak resistance |
The practical question is not whether Grok 4.7 is newer. It is whether the change improves accepted-task quality, speed and cost in the exact workflow. Keep prompts, tools, source material and scoring criteria fixed when comparing it with Grok 4.6.
Capabilities and best use cases
xAI recommends Grok 4.7 for both code and chat. The model is designed to work longer, manage extended context and verify its progress more carefully. Search tools are required for current events because the base model does not automatically receive real-time X or web data.
Capability | Practical use |
|---|---|
Coding | Repository work, debugging and long-running software-engineering tasks |
Knowledge work | Documents, presentations, legal tasks and professional analysis |
Long context | Up to 500,000 tokens with a higher price tier beyond 200K prompt tokens |
Search | Optional X Search and Web Search for current information |
Agent harnesses | Native Grok Bot awareness and compatibility with coding environments |
Safety controls | Updated refusal, jailbreak, cybersecurity and biology safeguards |
The strongest documented fit is coding, knowledge work and long-running tasks in the Grok ecosystem. A capable base model can still fail when retrieval is weak, tools are over-permissioned, instructions conflict or the system lacks verification. Evaluate the full application rather than the model in isolation.
Benchmark evidence
xAI reports 46.3% on CursorBench 4.0, 71.0% on DeepSWE v1.1 at high effort and 37.6% on Terminal-Bench 4.0. It also reports improvements on legal and clinical evaluations. The launch table uses xAI-selected configurations and should not be treated as an independent ranking.
Evaluation | Reported result | Evidence boundary |
|---|---|---|
CursorBench 4.0 | 46.3% | xAI-reported long-running coding result |
DeepSWE v1.1 | 71.0% at high effort | xAI-reported software-engineering result |
Terminal-Bench 4.0 | 37.6% | xAI-reported terminal-agent result |
Harvey Legal Agent Benchmark | 19.6% | xAI-reported legal-work result |
HealthBench Professional | 56.7% | xAI-reported clinical-reasoning result |
Benchmark scores depend on model version, reasoning effort, sampling, tools, scaffolding and the exact dataset release. Do not combine scores from different harnesses into a synthetic ranking. Use public results to choose candidates, then test those candidates on frozen tasks from the intended workflow.
Artificial Analysis Intelligence Index
Artificial Analysis independently measures model intelligence, speed and cost. Grok 4.7 scores 46 on the Artificial Analysis Intelligence Index at xhigh effort. Uses more than double the output tokens per task of Grok 4.6 at xhigh.
Metric | Value |
|---|---|
Intelligence Index score | 46 |
Effort setting | xhigh |
Source | Artificial Analysis (independent) |
Last verified | October 2, 2026 |
The Artificial Analysis Intelligence Index combines scores across mathematics, reasoning, coding, instruction following and language tasks. The scale is not a percentage: the index reflects relative position across the models they track, not a share of correct answers. Compare scores only within the same effort setting and the same index version.
Pricing and access
Grok 4.7 is available through the Grok API and xAI products, with additional access through selected development tools and platforms. A fast variant provides approximately twice the output speed at twice the standard token price.
The 200,000-token threshold matters for budgeting. Once the prompt reaches that level, xAI bills all tokens in the request at the higher long-context rates. Applications should compact histories, retrieve selectively and measure whether extra context improves accepted results.
Usage or access item | Current value |
|---|---|
Input below 200K prompt tokens | $2 per million tokens |
Cached input below 200K | $0.50 per million tokens |
Output below 200K | $6 per million tokens |
Input at or above 200K | $4 per million tokens |
Cached input at or above 200K | $1 per million tokens |
Output at or above 200K | $12 per million tokens |
Fast variant | Twice the standard output speed at twice the price |
Token price is only one part of total cost. Include cache behavior, long-context multipliers, tool calls, retries, wall time, failed-task recovery and human review. The useful comparison is cost per accepted result, not price per million tokens in isolation.
Limitations and deployment risks
- The 500,000-token context window is smaller than some competing million-token models.
- Long-context rates apply to every token once a prompt reaches 200,000 tokens.
- Current events require explicit X Search or Web Search tools rather than relying on the base model.
- The launch benchmark table is provider-reported and configuration-sensitive.
- Encrypted reasoning content is not a substitute for an auditable action log.
- Coding and cybersecurity tools require narrow permissions and authorization boundaries.
- Aliased model IDs can move to newer versions, so reproducible workflows should pin a dated snapshot when available.
High-stakes legal, medical, financial, scientific and security work needs qualified review. Log the exact model version and configuration, separate untrusted content from system instructions, restrict tools to the minimum required scope and require approval before irreversible or externally visible actions.
How to evaluate Grok 4.7
Build a frozen evaluation set of 20 to 50 real tasks. Include routine work, difficult edge cases, adversarial inputs, long-context examples and cases where the correct behavior is to stop or escalate. Compare Grok 4.7 with Grok 4.6 and at least one neighboring model using equivalent tools and source material.
Test area | What to record |
|---|---|
Task completion | Pass or fail against a written acceptance rubric |
Reliability | Repeated-run success rate, variance and silent failures |
Factuality | Unsupported claims, source use, quotations and citation accuracy |
Tool use | Wrong calls, retries, recovery and permission-boundary failures |
Long context | Retrieval accuracy, instruction retention and cost at realistic lengths |
Efficiency | Wall time, input, output, cached tokens, tool charges and review time |
Safety | Prompt injection, sensitive-data handling and irreversible-action controls |
Choose the least expensive configuration that clears the acceptance threshold with a safety margin. Re-run the suite when the model, system prompt, effort level, retrieval layer, tool definitions or approval policy changes.
Who should use it?
Situation | Recommendation |
|---|---|
Strong fit | coding, knowledge work and long-running tasks in the Grok ecosystem |
Pilot first | Long-running agents, large contexts, computer use and workflows with several tools |
Escalate | Ambiguous or consequential work that does not reliably clear the evaluation threshold |
Avoid unsupervised use | Irreversible actions, sensitive data or high-stakes decisions without monitoring and approval |
A migration should be driven by measured outcomes. Keep Grok 4.6 available during the pilot, record where each model succeeds or fails, and use routing when different task classes have different quality and cost requirements.
Frequently asked questions
Is Grok 4.7 open source?
No. Grok 4.7 is proprietary. Access, serving behavior and lifecycle decisions are controlled by xAI and supported distribution partners.
How much does Grok 4.7 cost?
$2 per million tokens. Review the full pricing table above because cached input, long context, processing mode and platform can materially change the total.
What is Grok 4.7 best used for?
Its strongest documented fit is coding, knowledge work and long-running tasks in the Grok ecosystem. Start with a supervised pilot and retain human sign-off for consequential work.
Should I migrate from Grok 4.6?
Only after a side-by-side evaluation. Measure accepted-task quality, total cost, latency, output style, tool reliability and migration engineering. A newer model is not automatically the better operational choice.
How should benchmark evidence for Grok 4.7 be interpreted?
Use provider results to identify promising workloads, then reproduce the comparison with the exact model configuration, tools and acceptance criteria that matter to the deployment. Do not infer a site ranking from benchmark claims gathered under a different harness.
Related model guides
Browse the Best AI Models directory.
Compare with the Grok 4.6 guide.
Compare with the Grok 4.5 guide.
Compare with the Gemini vs Grok comparison.
Official sources and update policy
xAI model and pricing documentation.
Checked October 2, 2026. We update this guide when xAI changes the specification, pricing, access, safety documentation or model lifecycle. Vendor benchmarks are attributed and are not presented as independent testing. No vendor payment or affiliate relationship determined inclusion.