SpaceXAI announced Grok 4.7 on 22 September, pitching it as its strongest model for coding and knowledge work - at the same price and the same speed as the model it replaces. ITmedia reports that the release moves to a larger base model and strengthens coding, long-running business tasks and safety, and that it is available the same day inside Cursor and through the API. The company is framing it as substantially cheaper and faster than comparable recent releases from rivals.
Key facts
| Fact | Detail |
|---|---|
| Question answered | Grok 4.7 Arrives for Coding and Knowledge Work, at the Same Price |
| Short answer | SpaceXAI announced Grok 4.7 on 22 September, keeping the previous model’s price and speed while moving to a larger base model and strengthening coding and long-running task performance. |
| Reading time | ~2 min |
| Last updated | 2026-09-22 |
Three details that matter more than the headline number
- The price held. A capability bump that does not move the meter on cost is the useful kind for anyone running an agent on a subscription or an API budget.
- The speed held. For an agent, latency is not a comfort feature. Every extra second multiplies across the dozens of steps a single task takes.
- The base model got larger. That is where the “long-running task” claim comes from, and it is the part worth testing rather than trusting.
Why “long-running tasks” is the interesting claim
A chat answer is a snapshot. An agent task is a sequence: read the folder, plan, act, check, retry. Benchmarks mostly measure the snapshot, which is why a model can look excellent on a leaderboard and still be frustrating on a desktop - it loses the thread, repeats a failed approach, or quietly drops a constraint that was set ten steps earlier.
Long-horizon reliability is also where cost actually lands. If step twelve is wrong, the first eleven were paid for and thrown away. That is why a model that is slightly less impressive in a single answer, but steadier across thirty steps, usually wins in practice.
If you run an agent on your PC
- Cheaper models make longer loops affordable. The practical ceiling on an agent is often budget, not ability.
- Match the model to the step. Planning, drafting and judging rarely need the same model, and price is now a legitimate input into that decision.
- A new model does not change the guardrails. Whatever the model, keep approvals in front of actions that write, delete, send or spend, and keep a log you can read back. Model quality reduces mistakes; it does not remove the need to see them.
MOOGH runs on Windows as a desktop agent that keeps you inside that loop: it plans first, shows the files and commands it wants to touch, and executes only after you approve. Download MOOGH and try the new models where they actually get used - on your own machine.
Related reading: