By xAI’s Grok 4.6 landed in GitHub Copilot on August 14 — two days after the model’s general release. It’s now selectable from the model picker across eight development surfaces, covering every major Copilot plan. The pricing is roughly one-fifth of comparable alternatives. If you use Copilot daily and haven’t checked your model picker recently, now’s the time.
Where Grok 4.6 Shows Up
Grok 4.6 is available on all eight Copilot surfaces: VS Code, Visual Studio, Copilot CLI, the Copilot cloud agent, the GitHub Copilot app, JetBrains IDEs, Xcode, and Eclipse. That’s the broadest rollout GitHub has given any new model addition. It covers Copilot Pro, Pro+, Max, Business, and Enterprise plans.
Rollout is gradual, so if Grok 4.6 isn’t in your picker yet, it should arrive soon. For Business and Enterprise teams, there’s one extra step: org administrators must explicitly enable the model in Copilot settings. It’s off by default. Developers won’t see it in the picker until an admin flips that switch.
What Grok 4.6 Is Built For
The model’s design focus is agentic coding — specifically, long-running tasks requiring sustained reasoning and tool use. xAI trained it with reinforcement learning on terminal coding, web development, kernel optimization, and CAD tasks. In practice, this means Grok 4.6 does well at:
- Multi-file refactors that require generating tests and iterating
- Visual and interactive projects where it can establish structure and design language in a single pass
- Knowledge work over complex, messy source material
- Extended coding sessions where output token cost becomes a factor
The 500K context window stays the same as Grok 4.5, and the model supports text plus image input. According to xAI’s release notes, training combined curated reasoning data with supervised fine-tuning and reinforcement learning across agentic environments.
Where It Trails
Terminal-heavy autonomy is the weak spot. On Terminal-Bench v3.0, Grok 4.6 scores 26% — a real improvement over Grok 4.5’s 15.7%, but still behind GPT-5.6 Sol’s 34.6%. For shell-centric agent workflows, that gap matters. On DeepSWE v1.1, the picture is more competitive: Grok 4.6 reaches 65.9% (up from 54%), while GPT-5.6 Sol leads at 73%.
The honest read: Grok 4.6 is strong for long reasoning sessions and structured coding tasks, and noticeably weaker for autonomous terminal execution compared to the top alternatives.
The Pricing Gap Is Significant
Under Copilot’s usage-based billing, Grok 4.6 runs at $2 per million input tokens and $6 per million output tokens. Compare that to GPT-5.6 Sol at $5 input / $30 output, or Claude Opus-class models at $5 input / $25 output. For agentic tasks that generate thousands of output tokens per session, Grok 4.6 is dramatically cheaper — not a rounding error, but a 4-5x difference on output.
| Benchmark | Grok 4.6 | GPT-5.6 Sol Max | Fable 5 Max |
|---|---|---|---|
| AA Intelligence Index | 61 | 61 | 62 |
| CursorBench v3.2 | 69.9% | 67.2% | 70.5% |
| DeepSWE v1.1 | 65.9% | 73% | — |
| Terminal-Bench v3.0 | 26% | 34.6% | — |
The Bigger Play
This isn’t just a model release. xAI is building distribution. Grok 4.6 is already available in Cursor, OpenRouter, Vercel, Cloudflare, and Grok Build — and now GitHub Copilot, the largest AI coding platform by user count at 26 million users. The strategy is clear: put the model wherever developers already work rather than pulling them to a new product.
Meanwhile, GitHub Copilot’s model picker has become the most consequential feature in the AI coding market. Claude Opus, Fable 5, GPT-5 variants, and now Grok 4.6 all live under one subscription. The era of picking an AI coding tool and sticking with it is over. The smarter move is picking the right model for each task — and GitHub just gave developers another strong option for long, complex agentic work.
The official GitHub Changelog entry has the full rollout details and feedback links. Check your Copilot model picker — if Grok 4.6 is there, it’s worth a test run on your next multi-file refactor.
