Grok 4.7: Specs, Context Window, Pricing, Benchmarks and Availability

Grok 4.7 is a proprietary large language model in xAI's Grok family, available on the xAI API since 21 September 2026 under the API identifier grok-4.7. The model page describes it as a frontier model for coding, agentic tasks and knowledge work. xAI's documentation and announcement now use the name SpaceX Super Intelligence (SpaceXSI); this page keeps the Grok and xAI names used in the API. It collects the published facts on one page: specifications, context window, prices including long-context rates, where the model runs, restricted variants, provider-reported benchmarks, data handling and retirement status. It is written for developers, architects and buyers who need the numbers with their sources, all from xAI's own documentation, pricing page, release notes and announcement.

Last checked: 2026-11-15.

This is an independent reference page from NHI Mgmt Group based on public provider documentation; it is not sponsored or endorsed by any provider.

Key takeaways

• On the xAI API since 21 September 2026. Grok 4.7 followed Grok 4.6 (12 August 2026) and is served at the same price as Grok 4.6, according to the announcement.
• A 500,000-token context window. It accepts text and image input, returns text, and has a stated knowledge cutoff of May 2026.
• Prices step up at 200,000 prompt tokens. Below that, $2 input, $0.50 cached input and $6 output per million tokens; at or above it, $4, $1 and $12 apply to every token in the request.
• Grok 4.7 Fast is not on the public API. The release notes say it is the same model at twice the token rates, offered only in Cursor and Grok Build.
• No model card or maximum output figure is published. xAI's model page is a specification page; it does not link a model card or system card or state an output token limit.

What is Grok 4.7?

xAI's model page describes Grok 4.7 as "SpaceXSI's frontier model for coding, agentic tasks, and knowledge work". The 21 September 2026 announcement calls it the provider's most capable model for coding and knowledge work, and states that it uses a larger new base model than Grok 4.6 with a longer reinforcement learning run on harder tasks, many of them taking hours. The announcement also says it is trained to understand the Grok Bot harness natively. For the full xAI catalogue see the xAI Models hub.

Grok 4.7 is a reasoning model with configurable effort. xAI publishes no parameter count and no open weights. The models page states that the plain model name points to the latest stable version, the -latest suffix to the latest version, and a dated suffix pins a release; the Grok 4.7 model page lists no other aliases.

Key facts at a glance

As of 2026-11-15.

Release and version history

The announcement describes the change from Grok 4.6 as a larger new base model plus a longer reinforcement learning run on harder, multi-hour tasks, rather than a new product line. Grok 4.6 itself arrived with the same 500k context window, pricing tiers and reasoning effort options that Grok 4.7 now lists. The models page was last updated on 21 September 2026. xAI publishes no dated snapshot identifier for Grok 4.7 on the pages checked.

How much does Grok 4.7 cost?

As of 2026-11-15, USD per million tokens, from the xAI pricing page. Once a prompt reaches 200,000 tokens, the long-context rates apply to all tokens in that request.

• Tool calls: Web Search $5 per 1,000 calls; X Search $5 per 1,000 posts and $10 per 1,000 profiles; Code Execution $5 per 1,000 calls; File attachment search $5 per 1,000 calls; Collections Search $2.50 per 1,000 calls; Remote MCP is token-based only.
• Storage: files $0.025 per GiB per day and collections $0.10 per GiB per day.
• Blocked requests: requests blocked by the usage guidelines before generation on the Responses API incur a $0.05 fee.
• Consumer apps: Not stated by xAI on the pages checked (2026-11-15); the announcement offers a free trial through Grok Build.

Worked cost examples

Illustrative arithmetic from the published rates above, before tool calls and without caching. Actual bills depend on reasoning tokens, which count as output.

The step at 200,000 tokens means a 300,000-token prompt costs about five times the 100,000-token example, not three times. The announcement states that Grok 4.7 is served at the same price and speed as Grok 4.6, and that the fast variant gives twice the output speed at twice the price.

For prices across providers see Frontier AI Models Compared.

Context window and output limits

Images and text can appear in any order in a request, according to the models page. Higher rate limits are requested through xAI sales. Although the window holds 500,000 tokens, every request at or above 200,000 prompt tokens is billed at the long-context rate for all of its tokens, so retrieval and summarisation designs that keep prompts under that line change the cost profile materially. See Context Windows Explained.

Modalities and tools

Because the Responses API always returns encrypted reasoning content for this model, applications that store conversation state should plan for that field. See Reasoning Models Explained and, for coding agents, Computer-Use and Browser Agents.

Where is Grok 4.7 available?

For general LLM API security set-up see the LLM Security Guide.

Access restrictions

• Standard model: generally available on the public xAI API; no allowlist stated.
• Grok 4.7 Fast: offered only in Cursor and Grok Build, not on the public API, per the release notes.
• Red-team capabilities: the announcement states that select cybersecurity partners have invite-only access to the model's red-team capabilities for defence research.
• Higher rate limits: by request to xAI sales.

Benchmarks (provider-reported, dated)

All scores below are reported by xAI in the Grok 4.7 announcement of 2026-09-21, at xHigh effort unless noted. Benchmark versions follow the provider's wording; see How to Read LLM Benchmarks.

The model page itself lists no benchmark scores.

Model card and system card

Not stated by xAI on the pages checked (2026-11-15): neither the Grok 4.7 model page nor the announcement links a model card or system card. The model page is the specification document, covering identifier, context, modalities, features, prices, rate limits and region. The announcement states that the model was built with a new safeguard stack, which xAI describes as its strongest tested so far on refusals and jailbreak resistance. See Model Cards and System Cards Explained.

See the Frontier AI Model Risk Frameworks Guide for how such reports are structured.

Data handling

• Training use: xAI's security FAQ states that API inputs and outputs are not used for training without explicit permission.
• Default retention: API requests and responses are stored for 30 days for auditing suspected abuse or misuse, encrypted at rest, then deleted automatically.
• Zero data retention (ZDR): enabled team-wide by a team admin in the Console under Team Settings, after existing Files and Collections are deleted. Prompts and outputs are then not persisted to disk. ZDR disables the stateful Responses API, Files, Collections, the Batch API, deferred completions and per-key request logging.
• Verification: an "Active" badge in Team Settings or the x-zero-data-retention response header confirms ZDR.
• Compliance: the FAQ states SOC 2 Type 2 compliance and offers a Business Associate Agreement questionnaire for HIPAA use.
• End-user attribution: since 25 September 2026 requests can carry safety_identifier, an opaque end-user identifier set by the application, so policy violations can be attributed to individual users; the legacy user field is still accepted (release notes).
• Custom moderation controls: since 27 October 2026 text API requests can include a guidelines array that sets how the model handles nsfw, violence and profanity content (release notes).

See Zero Data Retention for AI.

Deprecation status

Grok 4.7 is current and listed on the pricing page. No deprecation or retirement notice for Grok 4.7 appears in the release notes or on the models page as of 2026-11-15. The models page links retirement guides dated 15 May 2026 and 2 November 2026 that concern other models. Earlier Grok 4 versions (4.6, 4.5, 4.3 and 4.20) remain priced alongside it. Check the release notes before planning, because retirement notices are published there. See the AI Model Retirement Guide.

How Grok 4.7 differs from other models in its family

As of 2026-11-15. Models listed by release date. Sources: xAI pricing page and release notes. Prices are per million tokens below 200,000 prompt tokens.

• Grok 4.6: same context window, prices and reasoning effort options per the release notes; the announcement states that Grok 4.7 uses a larger new base model with a longer reinforcement learning run.
• Grok 4.3 and 4.20: 1M context windows and lower per-token prices; Grok 4.3 has no release-note date.
• US endpoint: only Grok 4.7 and Grok 4.6 are served on the US regional endpoint.

See the xAI Models hub for every version.

Practitioner checklist

• Track prompt sizes so requests stay below the 200,000-token threshold where long-context rates would double cost.
• Use the Batch API for non-urgent jobs, noting that ZDR disables it.
• Choose a default reasoning effort per workload; high is the default and xhigh is available.
• Store the reasoning.encrypted_content field if you maintain Responses API conversation state.
• Decide on ZDR before building on Files, Collections or the stateful Responses API.
• Budget tool calls separately from tokens, especially X Search, which bills per item returned.
• Confirm the US regional endpoint is needed before paying its 1.1x rate.
• Store xAI API keys in a secrets manager and rotate them; never embed them in client code.

Standards and references

• xAI: Grok 4.7 model page (accessed 2026-11-15)
• xAI: Release notes (accessed 2026-11-15)
• xAI: Pricing (accessed 2026-11-15)
• xAI: Models (accessed 2026-11-15)
• xAI: Grok 4.7 announcement (accessed 2026-11-15)
• xAI: API security FAQ (accessed 2026-11-15)

Related NHI Mgmt Group resources: xAI Models, Frontier AI Models Compared, Reasoning Models Explained, Zero Data Retention for AI.