Grok 4.7
SpaceXAI's flagship. Index 46, in the same band as MiMo-V2.6-Pro and behind Opus, Sonnet, Sol, Astra, and Argon. The independent coding-agent index moved up with this release. SpaceXAI's own launch table is not mixed into the scores on this site.
Every published effort
| Setting | Index | Per task | Tokens / sec |
|---|---|---|---|
| Extra high | 46 | $3.74 | 73 |
| High | 46 | $2.73 | 76 |
| Low | 42 | — | — |
What this workload costs
List price for one call. Reasoning tokens count as output. A cache hit uses the lab’s cache rate when one is published.
Specs
- API id
- grok-4.7
- Released
- 21 Sep 2026
- Context
- 500k
- Max output
- —
- Input / 1M
- $2.00
- Output / 1M
- $6.00
- Cache read / 1M
- $0.50
- Knowledge
- May 2026
- Inputs
- Text, Image
- Weights
- Closed
Rates double once the prompt reaches 200k, for the whole request. A prompt-cache key is how SpaceXAI says you actually hit the cache. US regional endpoint is 1.1×. Grok 4.7 Fast is the same model at about 2× speed and 2× price, in Cursor and Grok Build only. Docs say there is no separate text output cap.
Text and image in, text out.
Use it when
Coding agents and knowledge work in the Grok stack, especially if you want X search or the Grok Build harness. High and extra-high land on the same index score, so pay for high.
Leave it when
Extra-high by habit. It costs more per measured task here and does not move the index. Also a poor fit if you need a 1M window. Grok stops at 500k, and the last 300k of that is billed at double.
Tools named by the lab
- Function calling
- Web search
- X search
- Code execution