xAI’s Grok 4.6 is now rolling out in GitHub Copilot, two days after the model’s August 12, 2026 release, the company announced on August 14, 2026. The model — xAI’s current coding flagship, pitched at long-running agents and multi-step work — is selectable from Copilot’s model picker across eight surfaces: VS Code, Visual Studio, the Copilot CLI, the Copilot cloud agent, the Copilot app, JetBrains IDEs, Xcode, and Eclipse.
GitHub’s changelog entry, published the same day, confirms the rollout covers Copilot Pro, Pro+, Max, Business, and Enterprise plans. That spread is unusually wide for a new model addition: JetBrains, Xcode, and Eclipse support puts Grok 4.6 in front of developers well outside the VS Code core where most Copilot model launches concentrate.
GitHub said that in its internal testing, Grok 4.6 performed especially well on longer-horizon tasks requiring sustained reasoning and tool use across terminal-based coding in VS Code and the Copilot CLI. That aligns with how xAI framed the model at launch. “Grok 4.6 builds on Grok 4.5 with a particular focus on long-running agents and more ambitious interactive and visual work,” the company wrote in its release post.
What Grok 4.6 Brings to the Model Picker
Grok 4.6 arrives in Copilot with a set of vendor-reported evaluation results that place it near the top of the current coding class, with the caveat that applies to all self-reported numbers: these are xAI’s runs, on xAI’s chosen harnesses. In the company’s published eval table, Grok 4.6 High scores 61 on the Artificial Analysis Intelligence Index, a composite of nine benchmarks, matching OpenAI’s GPT-5.6 Sol Max and sitting one point behind Anthropic’s Fable 5 Max at 62. On GDPVal-AA, a knowledge-work evaluation, Grok 4.6 posts 1753 against 1728 for GPT-5.6 Sol Max and 1741 for Fable 5 Max.
The spread is wider on agentic coding benchmarks. Grok 4.6 scores 69.9% on CursorBench v3.2, ahead of GPT-5.6 Sol Max at 67.2% and behind Fable 5 Max at 70.5%. On DeepSWE v1.1 it reaches 65.9%, well ahead of its predecessor Grok 4.5 High at 54% but behind GPT-5.6 Sol Max’s 73%. On Terminal-Bench v3.0, which measures exactly the kind of terminal work GitHub highlighted, Grok 4.6 scores 26% against 34.6% for GPT-5.6 Sol Max, while still nearly doubling Grok 4.5 High’s 15.7%.
The model’s training recipe, per xAI’s account, paired a longer supplemental training run on curated reasoning and engineering data with supervised fine-tuning trajectories regenerated by Grok 4.5, followed by reinforcement learning across agentic tasks including kernel optimization, web development, and computer-aided design. The generation-over-generation jump on terminal and agentic benchmarks is the visible result, even if Grok 4.6 does not lead the category on every measure.
The Fine Print on Access and Billing
Two constraints shape who actually sees the model and what it costs. First, the rollout is gradual: GitHub notes that developers who don’t see Grok 4.6 in the picker yet should check back as availability expands. Second, on Copilot Business and Enterprise plans the Grok 4.6 policy is off by default, and an administrator must enable it in Copilot settings before seats can select the model.
On billing, Grok 4.6 is charged at provider list pricing under usage-based billing rather than drawing down a flat request allowance. Grok 4.6 is billed at provider list pricing; xAI lists API pricing starting at $2 per million input tokens and $6 per million output tokens, the same base rate GitHub charges for Grok 4.5 in its pricing table, and it puts Grok 4.6 well below the other flagship options in the picker on price: GPT-5.6 Sol lists at $5 input and $30 output per million tokens, and Claude Opus-class models at $5 and $25.
Outside Copilot, Grok 4.6 remains available through xAI’s API, Cursor, Grok Build, and partners including OpenRouter, Vercel, and Cloudflare. As covered in Unite.AI’s report on the launch, xAI offered doubled included usage in Grok Build and Cursor for the model’s first week, a window that closes August 19, 2026.
The Copilot integration extends a distribution pattern that has defined Grok 4.6’s first 48 hours: launch in xAI’s own tooling and Cursor, then arrive inside the largest third-party coding assistant by seat count. With the Business and Enterprise policy gate now in administrators’ hands, the pace of the model’s enterprise uptake will show up in Copilot settings before it shows up in any leaderboard.
Credit: Source link
