# Grok 4.7 Holds Grok 4.6 Prices and Moves the Specialty Benches

Times of AI Desk · 2026-09-21 · Models

[https://timesof.ai/2026/09/xai-grok-4-7](https://timesof.ai/2026/09/xai-grok-4-7)

> xAI shipped Grok 4.7 at the same $2/$6 token prices as Grok 4.6, with a fast variant at 2× speed and 2× price. Lab tables lead EEBench and Harvey’s legal-agent bench; Andon’s Vending-Bench put it in the middle — closer to Opus 5.5 than to Sol.

The price did not drop against Grok 4.6. The movement is on long coding, electrical engineering, and a legal-agent bench where the lab’s own numbers put it ahead of Sol and Fable. The independent cash test a few days later put it in the middle, closer to Opus 5.5 than to Sol.

**xAI** (September 21): **Grok 4.7**, which the lab calls its most capable model for coding and knowledge work. It uses a **larger base model** than Grok 4.6, a longer reinforcement-learning run weighted toward tasks that take many hours, and training to understand the **Grok Bot** harness natively.

## Price and availability

Starting at **$2 per million input tokens** and **$6 per million output tokens** — the same sticker as Grok 4.6. A **fast variant** serves at **twice the output speed and twice the price**. The hero line says “twice as fast, at half the price of comparable models.” The body says it is **served at the same price and speed as Grok 4.6**. Read the half-price claim against peers (GPT-5.6 Sol at $4 / $20, Fable 5.1 at $10 / $50 on the same table), not as a cut versus 4.6.

**Where**: **Cursor**, **Grok Build**, the **Grok API**, third-party coding harnesses, and model routers and cloud platforms, the day of the post.

## Lab table

xAI’s comparison: Grok 4.7 vs Grok 4.6 / GPT-5.6 Sol / Fable 5.1:

Effort settings differ: xAI generally uses Grok 4.7 at xhigh, Grok 4.6 at high, and competitors at max; the 71.0% DeepSWE result uses high. These settings limit like-for-like comparisons.

| Bench | Grok 4.7 | Grok 4.6 | GPT-5.6 Sol | Fable 5.1 |
| --- | --- | --- | --- | --- |
| CursorBench 4.0 | 46.3% | 40.4% | 41.7% | 51.8% |
| DeepSWE v1.1 | 71.0% (high) | 65.2% | 72.7% | 70.0% |
| EEBench | 64.0% | 53.0% | 39.4% | 56.4% |
| AA Briefcase v1.1 | 1,657 | 1,546 | 1,487 | 1,678 |
| Terminal-Bench 4.0 | 37.6% | 20.3% | 37.3% | 57.9% |
| Harvey Legal Agent | 19.6% | 15.8% | 2.5% | 6.7% |
| HealthBench Professional | 56.7% | 48.5% | 60.5% | 62.1% |

**Safety, lab-reported**: a new safeguard stack. xAI says it tops **LatchBio’s biosafety benchmark at 62.4%**, and on **HackerBench v0.3** lets **3.3%** of risky dual-use cyber prompts through while rarely blocking legitimate security work. Select cybersecurity partners get invite-only red-team access.

## Independent checks

- **Andon Labs**, posted **September 24, 2026**, ran Vending-Bench 2 (six runs, $500 and one simulated year). **Grok 4.7 averaged $10,537**. **GPT-6 Sol averaged $14,428**. **Claude Opus 5.5 averaged $9,235**. Andon says this is the first time a Grok model beat the newest Claude Opus on that bench, and that Opus 5.5 made less than Opus 5. In the four-game arena among the three, Sol finished well ahead and the other two ended within $200.

## Limits

- Price held flat versus Grok 4.6; “half the price” is a peer comparison.
- This article does not establish an Arena Code Elo or a confirmed Artificial Analysis index score for Grok 4.7.
- Biosafety and HackerBench figures are lab-reported.

**Correction and source review — October 14, 2026:** Removed an unpreserved historical Arena snapshot and an unverified secondary AA score. No Arena or AA ranking is inferred from their removal; the independent Vending-Bench results remain. Added the differing effort settings used in the vendor benchmark table.

## Sources

- [xAI: Introducing Grok 4.7 (September 21, 2026)](https://x.ai/news/grok-4-7)
- [Andon Labs: Opus 5.5, GPT-6 Sol and Grok 4.7 on Vending-Bench (September 24, 2026)](https://andonlabs.com/blog/opus-5-5-gpt-6-sol-grok-4-7-vending-bench)