xAI released Grok 4.7 on September 21, 2026, as its most capable model for coding and knowledge work. It is available on the xAI API as grok-4.7 at $2 per million input tokens and $6 per million output tokens for prompts under 200,000 tokens, the same price as Grok 4.6, with a 500,000-token context window. A faster variant exists, but the developer documentation says it is offered only in Cursor and Grok Build, not on the public API.
The company's website and documentation now use the name SpaceXAI; this article uses xAI, the name of its developer platform and API domain.
In its announcement, xAI says Grok 4.7 uses a new, larger base model than Grok 4.6, trained with a longer reinforcement learning run weighted toward problems that take many hours to complete. It describes three practical improvements: working longer on difficult tasks, checking its own work more carefully, and handling longer context. xAI also says it trained the model to understand its own Grok Bot harness, which it credits for better conversational and knowledge-work results.
The xAI release notes entry for September 21, 2026 adds the technical specifics:
reasoning.encrypted_content, even when the request does not ask for it.| Per million tokens (USD) | Prompt under 200K tokens | Prompt above 200K tokens |
|---|---|---|
| Input | $2.00 | $4.00 |
| Cached input | $0.50 | $1.00 |
| Output | $6.00 | $12.00 |
Source: xAI release notes, September 21, 2026 entry, checked October 10, 2026.
Grok 4.7's $6 output rate is below the $10 charged by GPT-6.1 Sol (per OpenAI's model page) and Claude Sonnet 5.5 (per Anthropic's pricing documentation). For output-heavy tasks such as code generation and long documents, that difference compounds. Cached input is a different story: at $0.50 per million tokens, it costs five times the $0.10 cached-input rate of GPT-6.1 Sol and the $0.10 cache-read rate of Sonnet 5.5, from the same two pages. The GPT-6.1 Sol guide and Claude Sonnet 5.5 guide cover those models in detail. Agents that re-read large cached contexts on every turn can therefore see the price advantage shrink or reverse. The right comparison is total cost per completed task on your own traffic mix.
The long-context step at 200,000 prompt tokens doubles all three rates, so the cheap tier covers the first 40% of the context window.
Grok 4.7 Fast is the same model at higher speed. The announcement describes it as twice the output speed at twice the price. The release notes add that it costs 2x standard rates, or 1.5x for long-context requests, and that it is available only in Cursor and Grok Build, billed through the plan there, not on the public xAI API. Teams planning to call the fast variant from their own code should not assume it is available.
Beyond the API, xAI says Grok 4.7 is available in Cursor, in its Grok Build coding agent, through third-party coding harnesses, and through model routers and cloud platforms.
xAI's announcement includes a table comparing Grok 4.7 at xhigh effort (except DeepSWE v1.1, where xAI marks Grok 4.7's 71.0% as a high-effort score) with Grok 4.6, GPT-5.6 Sol and Claude Fable 5.1 on CursorBench 4.0, DeepSWE v1.1, EEBench, AA Briefcase v1.1, Terminal-Bench 4.0, the Harvey Legal Agent Benchmark and HealthBench Professional. In xAI's numbers, Grok 4.7 leads on the electrical-engineering and legal benchmarks, trails Fable 5.1 on terminal work and clinical reasoning, and sits between the two OpenAI and Anthropic models on most coding tasks. xAI's headline claim is price-performance on CursorBench 4.0 rather than top score.
Claude Opus 5.5 (September 22), Claude Sonnet 5.5 (September 28) and GPT-6.1 Sol (September 29, per OpenAI's news feed) were released after Grok 4.7, so they are not in xAI's comparison. These are xAI's measurements, and this article does not reproduce them.
On safety, xAI says Grok 4.7 has a new safeguard stack, reports a 62.4% score on LatchBio's biosafety benchmark, and says it let 3.3% of risky dual-use prompts through on its own HackerBench v0.3. It also says select cybersecurity partners have invite-only access to the model's red-team capabilities for defensive research.
Teams on Grok 4.6 get a newer model at the same price. Re-test effort settings, since high is the default and xhigh is available.
Output-heavy workloads, such as code generation, report drafting and document creation, benefit most from the $6 output price.
Teams already using Cursor or Grok Build have access to both the standard and fast variants without separate API integration.
grok-4.7.reasoning.encrypted_content in Responses API payloads.When was Grok 4.7 released?
September 21, 2026.
How much does Grok 4.7 cost?
$2 input, $0.50 cached input and $6 output per million tokens for prompts under 200,000 tokens, and $4, $1 and $12 above, per xAI's release notes (checked October 10, 2026).
What is the context window of Grok 4.7?
500,000 tokens.
Can I use Grok 4.7 Fast through the API?
Not as of October 10, 2026. xAI's release notes say the fast variant is available only in Cursor and Grok Build.
Discover more content: