Grok 4.7 lands in Copilot — enabled by default
Grok 4.7 ships at the same price as 4.6 and rolls into GitHub Copilot, where new models are on by default until an admin turns them off. Check your model policy.
A new frontier model reached your developers' IDEs this week, and nobody on your team had to approve it. xAI released Grok 4.7 on September 21, and the same day GitHub added it to Copilot across Pro, Pro+, Max, Business and Enterprise. GitHub's own note is the part worth reading twice: new models are enabled by default unless an administrator disables them.
What actually happened
Grok 4.7 is xAI's latest reasoning model, pitched at agentic coding and long multistep work. It is selectable from the model picker in VS Code, Visual Studio, JetBrains, Xcode, Eclipse, the Copilot CLI, cloud agents and the mobile app. Rollout is gradual.
The price did not move. xAI's pricing page lists Grok 4.7 at $2.00 per million input tokens, $0.50 cached, and $6.00 output below 200k prompt tokens — identical to Grok 4.6. Cross the 200k threshold and every token in that request bills at the long-context rate: $4.00 in, $12.00 out. Context window is 500k. A separate Grok 4.7 Fast variant, available in Cursor and Grok Build, runs at double the standard rate.
The capability gain is real but not a clean sweep. Per benchmarks reported at launch, Grok 4.7 scores 46.3% on CursorBench 4.0 against 40.4% for 4.6 and 41.7% for GPT-5.6 Sol — but 51.8% for Fable 5.1. On DeepSWE v1.1 it posts 71.0%, ahead of 4.6's 65.2% and roughly level with the field. On Terminal-Bench 4.0 it jumps to 38.0% from 20.3%, still well behind Fable 5.1 at 57.9%.
Why your Copilot model policy matters for your business
Same price, better model is a good trade — we said as much when 4.6 landed. The problem is not the model. It is that the default is "on."
If you run Copilot Business or Enterprise, your model roster expands without a change request. That is fine until it isn't: a model with different data handling, a different refusal profile, and a different bill per token is now one dropdown away from your production code. Usage-based billing means the token cost lands on your invoice, not GitHub's.
Four things to do this week:
Open the model policy. Copilot settings, Business and Enterprise plans. Decide which models your org allows on purpose, rather than inheriting whatever shipped Monday. A short allowlist beats a long default.
Price the 200k cliff. Grok 4.7's long-context tier doubles input and output for the entire request, not the overflow. An agent that dumps a monorepo into context is not 10% more expensive — it is 2x. Cap what your agents can read.
Pin models in anything automated. IDE pickers are a human choice. CI jobs, cloud agents and scripted pipelines should name a model explicitly. "Latest" is not a version.
Benchmark on your repo, not theirs. CursorBench and Terminal-Bench are directional. The question that matters is whether a Grok 4.7 pull request passes your review at a lower total cost than the model you use now — and that is a week of real tickets, not a leaderboard.
Key takeaways
- Grok 4.7 released September 21 and reached GitHub Copilot the same day, across Pro through Enterprise SKUs
- GitHub enables new Copilot models by default; admins must disable them through model policy
- Pricing is unchanged from Grok 4.6 — $2.00/$0.50/$6.00 per 1M tokens below 200k, doubling to $4.00/$12.00 above
- The long-context tier reprices the whole request, so context bloat costs 2x, not 10%
- Benchmarks improve over 4.6 across CursorBench, DeepSWE and Terminal-Bench, but trail Fable 5.1 on two of three
- Pin model names in CI and agent pipelines; leave the picker to humans
Your AI coding stack should be a decision, not a default. We set model policy, context budgets and pinned versions so a vendor's Monday release doesn't quietly become your Tuesday bill. See how we work or tell us what your team is running.
Sources: GitHub Changelog, xAI pricing docs, Unite.AI.
- #grok
- #github-copilot
- #ai-coding
- #vendor-pricing
- #developer-tools
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
Intrinsic Core goes Apache 2.0 — the runtime runs locally
Alphabet's Intrinsic open-sourced its industrial robotics control stack under Apache 2.0 at ROSCon 2026. Why a local runtime is the detail that matters.
Read itUbuntu 26.10 moves cp, mv and rm to Rust coreutils
Ubuntu 26.10 completes the Rust coreutils switch on October 15. Your CI scripts and base images get new cp, mv and rm implementations — here's how to test first.
Read it