xAI releases Grok 4.7 with 500K context for coding and agents

xAI has released Grok 4.7 through its API for coding, agentic tasks, and knowledge work. The model accepts text and images, exposes a 500,000-token context window, and supports low, medium, high, and xhigh reasoning effort. That combination puts the release in the practical frontier-model lane: teams can test it against large repositories, long research packets, and multimodal workflows without waiting for a separate hosted rollout.

Grok 4.7 targets long-running software work

xAI describes Grok 4.7 as a frontier model for coding, agents, and knowledge work. The release notes list text and image inputs, text-only output, no text output limit, and availability on the US regional endpoint. The Responses API also returns encrypted reasoning content even when the request does not list it in include.

The model supports four reasoning levels: low, medium, high, and xhigh, with high as the default. That gives developers a direct latency-versus-depth control instead of forcing one fixed inference profile across every task. The practical test is whether higher effort improves repository-level planning enough to justify the extra cost and time.

Pricing changes at the 200K-token boundary

xAI lists one price tier below 200,000 prompt tokens and a higher tier above it:

The 500K context window therefore comes with a material cost step-up for long prompts. Teams evaluating the model should measure prompt size, cache reuse, and output length separately; a nominally cheaper model can become expensive when large repository context repeatedly crosses the 200K threshold.

Availability extends beyond the xAI endpoint

OpenRouter lists Grok 4.7 as an available model and reports a 500K context window, with a separate provider-routing layer and benchmark view. Its page identifies the model as succeeding Grok 4.6 and includes Artificial Analysis results, but those benchmark values are reported for xhigh reasoning and should not be treated as a universal score for every effort level or provider route.

For developers, the immediate access paths are the xAI API and compatible routing through OpenRouter. The model page also makes the operational trade-off visible: routing can improve provider resilience, while direct xAI access gives the clearest view of the vendor’s own pricing, region, and feature behavior.

What to verify in a first test

Start with a representative codebase or long technical document, then compare low, high, and xhigh reasoning on the same task. Track completion quality, latency, output tokens, cache hits, and whether the 200K prompt boundary is crossed. Image input should be tested separately because a text-only benchmark will not reveal the model’s multimodal workflow value.

The release is available now in the API, but the documentation does not by itself establish identical availability, limits, or pricing in every downstream product. Check the selected endpoint and provider before moving production traffic. For broader workflow patterns, see the AI workflow automation guide.

Sources and methodology

This report uses xAI’s developer release notes as the primary source and OpenRouter’s Grok 4.7 model page as independent availability and benchmark context. Benchmark figures on OpenRouter are presented as reported provider data, not as an independent LinkLoot test.