Grok 4.7 Is Here — but Not in the Consumer App Yet

SpaceXAI released Grok 4.7 on Sept. 21, 2026, calling it its most capable model for coding and knowledge work. It is available now in several developer tools, but it has not yet replaced the model in the consumer Grok experience on the web, mobile apps, or X.

So basically: Grok 4.7 is a developer-first release focused on longer coding and professional tasks. The consumer rollout will follow later.

What changed

SpaceXAI says Grok 4.7 uses a larger base model than Grok 4.6 and received a longer reinforcement-learning run on a harder mix of tasks. That work emphasized problems that take many hours to complete rather than short, isolated prompts.

The company also says the model is better at checking its own work and managing longer context. Those improvements are intended for agentic workflows in which a model reads a large repository, works through multiple steps, runs tools, and verifies the result before returning it.

The release is available now through:

• Cursor
• Grok Build
• the Grok API
• third-party coding harnesses, model routers, and cloud platforms

GitHub separately announced that Grok 4.7 is beginning a gradual rollout in GitHub Copilot. It will be offered to Copilot Pro, Pro+, Max, Business, and Enterprise customers across products including Visual Studio Code, Visual Studio, Copilot CLI, GitHub’s cloud agent, JetBrains, Xcode, and Eclipse. Business and Enterprise administrators can manage access through Copilot’s model policies.

Consumer Grok comes later

The release does not mean every Grok user has access to the new model today. SpaceXAI’s model card says Grok 4.7 will come to its consumer surfaces — the web, mobile apps, and Grok on X — at a later date.

That creates a split launch. Developers can use the model through supported tools and APIs now, while people using the regular Grok interface will continue waiting for a separate rollout. SpaceXAI did not give a date for that expansion in the materials published Sept. 21.

Pricing stays level with Grok 4.6

Standard Grok 4.7 pricing starts at $2 per million input tokens and $6 per million output tokens, matching the prices SpaceXAI lists for Grok 4.6. The company also offers a fast variant with twice the output speed at twice the price.

That makes the standard version a direct upgrade within the existing price tier, at least according to the company’s published rates. Actual costs in products such as Copilot may depend on each platform’s billing rules. GitHub says Grok 4.7 usage is billed at provider list pricing under its usage-based billing system.

The benchmark gains are company-reported

SpaceXAI highlights two coding-agent benchmarks:

• On CursorBench 4.0, which focuses on longer-running coding tasks, Grok 4.7 scored 46.3% at xhigh, compared with 40.4% for Grok 4.6.
• On Terminal-Bench 4.0, Grok 4.7 scored 38.0%, compared with 20.3% for Grok 4.6.

Both results come from SpaceXAI’s launch materials and model card. They indicate a sizable improvement over the previous Grok version under the reported test conditions, especially on terminal work, but they are not independent evaluations. Benchmark scores can also vary with the agent harness, reasoning setting, and evaluation setup.

The company describes Grok 4.7 as competitive with other frontier models on price and performance. That comparison should be read as a vendor claim until broader testing shows how the model performs across real repositories and repeated long-running tasks.

What the release means now

Grok 4.7 expands the model choices available to developers working in Cursor, Copilot, APIs, and other agent platforms. Its stated improvements target the part of coding-agent work that remains difficult: sustaining a task across many steps without losing context or skipping verification.

The next evidence will come from day-to-day use. Copilot customers may receive access at different times during the gradual rollout, and SpaceXAI has not said when the consumer Grok apps will move to 4.7.