The Vibe Father
Grok 4.6 pricing is the same headline as Grok 4.5, with two important changes underneath. SpaceXAI charges $2 per million input tokens and $6 per million output tokens when the prompt stays under 200,000 tokens. Cached input is $0.50, up from $0.30 on Grok 4.5. Prompts that reach 200,000 tokens double every cell.
That is still far below Claude Opus 5 at $5 / $25 and the $5 / $30 band GPT-5.6 Sol has used in public comparisons. Artificial Analysis measured about $0.84 per Intelligence Index task for Grok 4.6. Treat that as their harness cost, not your repository’s bill.
The price sheet that matters
| Prompt size | Input | Cached input | Output |
|---|---|---|---|
| Under 200k tokens | $2 | $0.50 | $6 |
| 200k tokens or more | $4 | $1 | $12 |
A fast variant is listed at twice those rates. Reasoning effort has four settings, low, medium, high, and xhigh, with high as the default. Higher effort usually means more output tokens, which is the expensive column. If you leave xhigh on for routine edits, the $6 line will dominate the invoice even when the input price looks cheap.
How to keep the bill honest
- Set a prompt cache key so multi-turn agent loops hit the same server. SpaceXAI says cache-cold retries often pay full input price.
- Use context compaction on long loops instead of resending the whole transcript.
- Reserve xhigh for the jobs that failed at high.
- Watch the 200k threshold. One oversized prompt reprices the entire request.
- Expire the first-week 2x usage in Grok Build and Cursor as a trial, not a permanent budget.
Access is not the same as a bargain
The model is on the SpaceXAI API, Grok Build, Cursor, OpenRouter, Vercel, and Cloudflare. A Cursor Ultra or SuperGrok Heavy seat is a product bundle, not a token-price rewrite. If you already bring your own key, the public sheet above is the one to model. If you only have included usage, measure remaining quota before you point a nightly agent at it.
Compare that sheet with when expensive models earn their keep and with the live scorecard. Price without a score is a rumor. A score without a price is a poster.
Bottom line
Grok 4.6 pricing is a same-sticker upgrade with a steeper cache line and a sharp long-context cliff. Staff it where agentic quality matters and output tokens are controlled. Do not assume the first-week bonus or the fast variant will look like $2 / $6 once the work is real.
