Elon Musk’s xAI, also referred to as SpaceXAI in some announcements, launched Grok 4.7 on September 21, 2026, positioning it as the firm’s most capable system yet for coding and extended knowledge-work tasks. The model became available immediately across the Grok app, Cursor, Grok Build, the xAI API, third-party coding harnesses, model routers, and cloud platforms, with no waitlist.

Company statements described Grok 4.7 as a notable improvement over its predecessor, Grok 4.6, delivered at the same price and speed. It employs a larger base model and underwent longer reinforcement-learning training focused on harder multi-hour problems. The system spends more time on difficult tasks, verifies its own outputs more thoroughly, manages extended context more effectively, and incorporates native understanding of the Grok Bot harness for conversational and general knowledge applications.

Technical Scale and Training Details

Grok 4.7 contains 2.1 trillion parameters, a 40 percent increase from the 1.5 trillion parameters in Grok 4.6. Supplemental training incorporated data from SpaceX operations, including Starlink satellite telemetry, manufacturing records, and engineering failure logs. This addition aims to strengthen reasoning about hardware and physical systems beyond text-based internet data alone.

The model supports a 500,000-token context window, accepts text and image inputs with text outputs, and offers adjustable reasoning effort levels of low, medium, high (default),and xhigh. It includes tools for function calling, web search, X search, and code execution. xAI stated that Grok 4.7 carries the company’s strongest safety guardrails to date, with improved performance on refusals and jailbreak resistance.

Pricing and Performance Claims

API pricing remains unchanged from Grok 4.6 at $2 per million input tokens and $6 per million output tokens for standard use under 200,000 input tokens. Rates rise for longer contexts. xAI characterized the model as twice as fast and half the price of comparable frontier systems. A faster variant is available in select environments at double the token rates and double the output speed.

Musk described the release on X as delivering a strong combination of intelligence, speed, and low cost. The company noted that the model is highly competitive in its class for price-performance, particularly on longer-running coding tasks measured by CursorBench 4.0.

Benchmark Results Across Coding and Knowledge Work

Published comparisons place Grok 4.7 ahead of Grok 4.6 across multiple evaluations and competitive with or ahead of certain rival models on specific tests, while remaining behind the leading systems overall on several others.

On GDPval, which evaluates economically valuable knowledge work such as legal memos, spreadsheets, and slide decks, Grok 4.7 scored 1,695, behind Claude Fable 5.1 at 1,735. Independent evaluations from Artificial Analysis recorded gains of more than 100 Elo points over Grok 4.6 on AA-Briefcase and solid advances on coding-agent indexes when paired with Grok Build.

Grok 4.7 leads the listed comparisons on electrical engineering and legal-agent tasks while trailing Fable 5.1 on CursorBench, multi-hour office work, terminal tasks, and health-related reasoning. Results against GPT-5.6 Sol are stronger in most categories except DeepSWE and HealthBench.

Release Timing and Competitive Context

The launch followed at least five public timeline adjustments by Musk since late July, including references to a four-week window, a few weeks, three to four weeks, ten days, and additional days of preparation. It arrives just over a month after Grok 4.6 and ahead of anticipated new releases from OpenAI and Anthropic.

OpenAI is reported to be preparing a successor to GPT-6 Astra, while Anthropic is considering a response model. Shares of SpaceX traded higher by nearly 2 percent near $154 around the announcement period, according to market data.

Musk had indicated before the release that Grok 4.7 would land roughly on par with Anthropic’s Claude Opus 5.0 rather than newer variants, with multimodal capabilities still under development. He outlined subsequent steps including Grok 4.8 as a further advance, Grok 4.9 targeting the class of current leaders, and Grok 5 as a potential frontier model, without providing firm dates.

Grok 4.7 is accessible now for users of the Grok interface, coding environments, and API developers seeking a combination of extended reasoning, hardware-oriented training, and cost efficiency relative to higher-priced competitors.