xAI launched Grok 4.7 on September 21, 2026, positioning it as a model for coding, knowledge work, long-running tasks and agent workflows. The model improves on Grok 4.6 across every benchmark listed in xAI’s comparison table, but it does not lead every test: Fable 5.1 Max remains ahead on several coding and long-task evaluations. API pricing starts at $2 per million input tokens and $6 per million output tokens.
Grok 4.7 availability and pricing
In its official announcement, xAI listed Grok 4.7 for Cursor, Grok Build, the Grok API, third-party coding harnesses, model routers and cloud platforms. That gives developers several ways to evaluate the model, from an editor-based workflow to direct API calls.
The standard API rates are:
- $2 per million input tokens
- $6 per million output tokens
xAI also described a fast variant that delivers twice the output speed at twice the standard price. These are token rates, so the amount charged for a complete task depends on how many tokens the workflow consumes and how much output it generates.
For additional context on how API pricing shapes model choice, see NeoTeo’s comparison of API costs in the wider model market.
What changes from Grok 4.6
xAI says Grok 4.7 uses a new, larger base model and a longer reinforcement-learning run focused on harder problems that can take many hours. Reinforcement learning is the stage in which a model is trained through feedback to improve how it handles selected tasks.
The company also says Grok 4.7 is better at checking its own work and managing longer context. Its stated design includes native understanding of the Grok Bot harness, the agent framework used to coordinate different roles within a project.
A demonstration of that workflow shows personas assigned to project-management, development and design roles. That makes the agent-oriented pitch concrete: Grok Bot can divide a project into separate roles and coordinate their work. The practical value still depends on the task and the surrounding workflow, rather than on the model name alone.
Where Grok 4.7 gains—and where rivals lead
The official xAI table compares Grok 4.7 xHigh with Grok 4.6 High, GPT-5.6 Sol Max and Fable 5.1 Max. The modes differ in name, so the figures should be read as the configurations listed by xAI rather than as a universal ranking of every version of each model.
| Benchmark or task | Grok 4.7 xHigh | Grok 4.6 High | GPT-5.6 Sol Max | Fable 5.1 Max |
| CursorBench 4.0 | 46.3% | 40.4% | 41.7% | 51.8% |
| DeepSWE v1.1 | 71.0% | 65.2% | 72.7% | 70.0% |
| EEBench | 64.0% | 53.0% | 39.4% | 56.4% |
| AA Briefcase v1.1 | 1,657 | 1,546 | 1,487 | 1,678 |
| Terminal-Bench 4.0 | 38.0% | 20.3% | 37.3% | 57.9% |
| Harvey Legal Agent Benchmark | 19.6% | 15.8% | 2.5% | 6.7% |
| HealthBench Professional | 56.7% | 48.5% | 60.5% | 62.1% |
Grok 4.7’s largest predecessor gain in the table appears on Terminal-Bench 4.0, where its 38.0% score is above Grok 4.6’s 20.3%. It also moves from 40.4% to 46.3% on CursorBench 4.0, from 53.0% to 64.0% on EEBench and from 15.8% to 19.6% on the Harvey Legal Agent Benchmark.
The model leads the listed competitors on Harvey and EEBench. Fable 5.1 Max leads Grok 4.7 on CursorBench, AA Briefcase, Terminal-Bench and HealthBench Professional, while GPT-5.6 Sol Max is higher on DeepSWE. The result is a workload-dependent profile, not an across-the-board victory.
What the benchmark split means for developers
For coding, the table points in two directions. Grok 4.7 improves substantially over Grok 4.6 on terminal work and performs above GPT-5.6 Sol Max on Terminal-Bench 4.0, but Fable 5.1 Max reaches 57.9% on the same test. On DeepSWE v1.1, Grok 4.7 scores 71.0%, below GPT-5.6 Sol Max at 72.7% and above Fable 5.1 Max at 70.0%.
For legal work, Grok 4.7 has the highest score among the four models in the table: 19.6% on the Harvey Legal Agent Benchmark. That result describes the listed benchmark condition; it does not turn the model into a substitute for professional legal judgment.
The same caution applies to the HealthBench Professional result. Grok 4.7 reaches 56.7%, while Fable 5.1 Max scores 62.1% and GPT-5.6 Sol Max scores 60.5%. A benchmark result can help identify a model worth testing for a workload, but it does not determine performance on every prompt, toolchain or production system.
Is Grok 4.7 cheaper per task?
Not necessarily. The listed API price and the cost of completing a task measure different things.
A separate Intelligence Index task-consumption figure associated with Artificial Analysis put Grok 4.7 at approximately 81,000 output tokens per task, compared with about 36,000 for Grok 4.6 and 27,000 for GPT-6 Astra. Because output-token usage varies, a lower price per million tokens does not automatically produce a lower bill for a completed task.
The practical calculation for a deployment therefore has two parts: the provider’s input and output rates, and the number of tokens the workflow actually uses. Long prompts, repeated tool calls, extended reasoning and large generated responses can all change the total.
Where developers can use Grok 4.7
xAI’s listed access points cover both direct development and agent-based work:
- Cursor for coding workflows
- Grok Build for building with Grok
- The Grok API for direct application integration
- Third-party coding harnesses for custom agent workflows
- Model routers for routing requests between models
- Cloud platforms that offer the model through their services
Grok 4.7 is therefore most interesting when the job matches its stronger results: terminal work, legal-agent evaluations, general professional tasks and workflows that benefit from the Grok Bot harness. The official comparison also gives developers a reason to test alternatives when their priority is terminal performance, long-form office work or clinical reasoning.
