Grok 4.7 in GitHub Copilot: Pricing, Benchmarks & 500K Context

Grok 4.7 Launch and Availability

On September 21, 2026, xAI officially released Grok 4.7, its latest reasoning model. The launch was accompanied by a rollout in GitHub Copilot, making Grok 4.7 available to users across all major Copilot subscription tiers: Pro, Pro+, Max, Business, and Enterprise. According to GitHub and corroborating sources, this rollout began on the same day as the model’s release. This simultaneous availability meant developers using GitHub Copilot could access Grok 4.7 immediately. In addition to GitHub Copilot, Grok 4.7 is live through several other channels. These include the Grok API, grok.com, the iOS and Android Grok apps, and Cursor, which provides access through its desktop, web, and iOS platforms, as well as command-line tools and SDKs. This broad distribution ensures that Grok 4.7 is accessible to a wide range of developers and organizations through multiple platforms.

Product Definition and Technical Capabilities

Grok 4.7 is described as xAI’s reasoning model optimized for coding and advanced knowledge work. According to xAI’s announcement materials and GitHub, it is designed for agentic coding and complex, multistep workflows. The model is intended to support scenarios where autonomous agents must reason through problems and handle extended chains of logic. Grok 4.7 builds on Grok 4.6 and is constructed on a new, larger base model, as stated by xAI. This architectural change is intended to enhance the model’s capabilities for advanced reasoning tasks. Grok 4.7 features a 500,000-token context window overall, which allows it to process extensive inputs and maintain context over longer interactions. This context window is particularly relevant for tasks that require deep reasoning and multi-step problem-solving, as it enables the model to reference a larger amount of information within a single session.

Pricing Structure for Grok 4.7

Grok 4.7’s pricing model is structured with multiple tiers. For prompts under 200,000 tokens, xAI charges $2 per million input tokens and $6 per million output tokens for API access. This pricing structure remains consistent with Grok 4.6, allowing developers to transition to the new model without incurring additional costs. For cached input, the rate is $0.50 per million tokens for prompts below the 200,000-token threshold. When prompts exceed 200,000 tokens, the pricing increases to $4 per million input tokens, $12 per million output tokens, and $1 per million cached input tokens. Grok 4.7 is also available through OpenRouter, which resells the model at $1.60 per million input tokens and $4.80 per million output tokens, with a $0.40-per-million cache-read rate and $5 per 1,000 web-search calls. A fast variant of Grok 4.7 is available, running at roughly double the output speed of the standard model. This variant is priced at $4 per million input tokens and $12 per million output tokens on the sub-200K tier. This tiered pricing structure provides flexibility for developers based on their usage needs.

The pricing for Grok 4.7 is positioned to undercut other leading models in the market. For example, GPT-5.6 Sol is listed at $4 per million input tokens and $20 per million output tokens, while Claude Fable 5.1 is listed at $10 per million input tokens and $50 per million output tokens. This means Grok 4.7’s input and output token prices are lower than those of these competitors. The fast variant, which offers double the output speed, is available at double the price of the standard model for prompts under 200,000 tokens.

Performance Benchmarks

Grok 4.7 has been evaluated on several benchmarks, providing a detailed view of its performance across different tasks. On DeepSWE v1.1, under the high reasoning-effort setting, Grok 4.7 scored 71.0%. This score edged out Claude Fable 5.1 Max at 70.0% and Grok 4.6 at 65.2%, while trailing GPT-5.6 Sol Max at 72.7%. This benchmark is designed to test reasoning and problem-solving capabilities in software engineering contexts. On CursorBench 4.0, Grok 4.7 scored 46.3%, ahead of GPT-5.6 Sol’s 41.7%. CursorBench 4.0 is used to evaluate agentic coding and workflow automation tasks. On EEBench, which covers electrical engineering tasks, Grok 4.7 reached 64.0%, surpassing GPT-5.6 Sol Max (39.4%) and Fable 5.1 Max (56.4%). This suggests Grok 4.7 performs well in technical domains requiring specialized knowledge.

Terminal-Bench 4.0 results for Grok 4.7 show some variability: one source reports a score of 26%, while another lists 38.0%. Both scores are lower than those for GPT-6 Astra (60%) and Claude Fable 5.1 (55% or 57.9%). Terminal-Bench 4.0 is designed to test command-line and system-level reasoning. On LatchBio’s biosafety benchmark, Grok 4.7 scored 62.4%. This benchmark evaluates the model’s ability to handle biosafety-related prompts.

On HackerBench v0.3, a benchmark for evaluating responses to potentially dangerous dual-use prompts, Grok 4.7 kept the pass-through rate for such prompts to 3.3%. This metric reflects the model’s handling of security-sensitive queries.

Composite and System-Level Metrics

According to the Artificial Analysis Intelligence Index (version 4.3.2), Grok 4.7 scores 46. This score places it below Claude Fable 5.1 and GPT-6 Astra, both of which score 53 according to Artificial Analysis. These composite scores are intended to provide a high-level comparison of model performance across a range of tasks. Artificial Analysis also puts Grok 4.7’s effective cost per task at $2.73 in its “high” reasoning-effort tier. This cost metric is calculated based on the model’s pricing and its performance on complex reasoning tasks.

Conclusion

Grok 4.7 is now broadly available through GitHub Copilot and other major distribution channels, with a tiered pricing structure and a large context window. Its benchmark results show competitive performance in several technical domains, and its pricing is positioned below other leading models. The model’s technical specifications and performance metrics are documented in both xAI’s official materials and third-party reporting.