xAI Unveils Grok 4.7: Same $2/$6 Pricing as Grok 4.6, Now With a Bigger Base Model

If you build with AI models for coding, automation, or knowledge-heavy work, the newest release from SpaceXAI is worth a close look. The company has launched Grok 4.7, its latest flagship model designed for software development, agent-style tasks, and professional work. This article walks you through what's new inside the model, how it performs against rivals like GPT-5.6 Sol Max and Fable 5.1 Max, what safety measures it includes, and what it costs to run — all so you can decide whether to switch or stay put.

The headline is simple: Grok 4.7 sits on a bigger foundation model and a much longer training run, yet it keeps the identical pricing and speed of Grok 4.6.

Ready to Use Today

You don't have to wait to try it. Grok 4.7 is available right now as a hosted model through the xAI API, Cursor, Grok Build, OpenRouter, Vercel, and Cloudflare. The model name to call is grok-4.7.

Key specifications from the developer docs:

• Context window: 500,000 tokens
• Knowledge cutoff: May 2026
• Modalities: text and image input, text output
• Reasoning effort levels: low, medium, high (default), xhigh
• APIs supported: Responses API and Chat Completions
• Tools: function calling, web search, X search, and code execution

What's New Compared to Grok 4.6

SpaceXAI highlights four upgrades over the previous version:

1. A new, larger base model. Grok 4.7 is not a re-tune of the old base; it starts from a bigger one entirely.
2. A longer reinforcement learning run on tougher problems. The training mix leans heavily toward tasks that demand many hours to finish.
3. Improved self-checking and long-context work. According to the company, the model reviews its own output more rigorously.
4. Built-in understanding of the Grok Bot harness. The model was trained natively on this harness for conversational and knowledge-work scenarios.

How It Scores on Benchmarks

The launch table pits Grok 4.7 (at xHigh effort) against Grok 4.6 High, GPT-5.6 Sol Max, and Fable 5.1 Max. Note that every score here is vendor-reported, so treat them as a starting point rather than gospel.

| Benchmark | Grok 4.7 xHigh | Grok 4.6 High | GPT-5.6 Sol Max | Fable 5.1 Max |
|---|---|---|---|---|
| Input price ($/M tokens) | $2 | $2 | $4 | $10 |
| Output price ($/M tokens) | $6 | $6 | $20 | $50 |
| CursorBench 4.0 | 46.3% | 40.4% | 41.7% | 51.8% |
| DeepSWE v1.1 | 71.0%* | 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 | 38.0% | 20.3% | 37.3% | 57.9% |
| Harvey Legal Agent Benchmark | 19.6% | 15.8% | 2.5% | 6.7% |
| HealthBench Professional | 56.7% | 48.5% | 60.5% | 62.1% |

*Run at high effort

Grok 4.7 beats Grok 4.6 on every row. The standout gains are on Terminal-Bench 4.0, which nearly doubled from 20.3% to 38.0%, and EEBench, which climbed 11 points to 64.0% — the best figure in that column. On the Harvey legal agent test, Grok 4.7's 19.6% far outpaces Fable 5.1 Max's 6.7%.

That said, it doesn't sweep the field. Fable 5.1 Max leads on 4 of 7 benchmarks, including a strong 57.9% on Terminal-Bench 4.0, and GPT-5.6 Sol Max holds the top DeepSWE v1.1 score at 72.7%.

Cost is where Grok 4.7 makes its case. Fable 5.1 Max charges roughly 5 times more per input token and about 8.3 times more per output token, while GPT-5.6 Sol Max runs about 2 times and 3.3 times more expensive respectively. SpaceXAI positions Grok 4.7 at the price-performance frontier on its CursorBench 4.0 cost-per-task chart.

For professional knowledge work, the GDPval evaluation gave Grok 4.7 xhigh a score of 1,695 Elo, up from 1,605 for Grok 4.6 high. Fable 5.1 Max still leads there at 1,735, and GPT-6 Astra Max scored 1,542. SpaceXAI also notes the model is stronger at producing documents and presentations.

Safety and Cybersecurity Changes

Grok 4.7 arrives with a completely rebuilt safeguard stack. SpaceXAI describes it as the most resistant model it has tested against jailbreaks and policy violations. On LatchBio's biosafety benchmark, it scored 62.4%, topping that test.

On HackerBench v0.3 — SpaceXAI's internal benchmark for risky and malicious cyber prompts — only 3.3% of risky dual-use requests got through. The company claims the model rarely blocks legitimate security work in the process. Select cybersecurity partners can also request invite-only access to red-team capabilities for defense-focused research.

Pricing, Availability, and Quick-Start Code

Grok 4.7 costs $2 per million input tokens and $6 per million output tokens. It works on every Cursor plan, serves as the default model in Grok Build, and is also served through the Grok API, OpenRouter, Vercel, and Cloudflare.

A faster variant, Grok 4.7 Fast, delivers double the output speed at double the price. According to the docs, it runs only inside Cursor and Grok Build — not via the public xAI API — and it's not part of Grok Build's free tier.

For teams with data-residency needs, a US regional endpoint at `https://us.api.x.ai/v1` keeps inference inside the United States for a 10% premium. SpaceXAI suggests setting a `prompt_cache_key` to get dependable cache hits.

Here's a minimal Python example from the docs:

```python
import os
from xai_sdk import Client
from xai_sdk.chat import user

client = Client(api_key=os.getenv("XAI_API_KEY"))
chat = client.chat.create(model="grok-4.7")
chat.append(user("Explain this repo."))
print(chat.sample().content)
```

Key Takeaways

• Pricing is unchanged from Grok 4.6: $2 input and $6 output per million tokens.
• Grok 4.7 tops the launch table on EEBench (64.0%) and the Harvey Legal Agent Benchmark (19.6%).
• Fable 5.1 Max still wins 4 of 7 benchmarks, including Terminal-Bench 4.0 at 57.9%.
• The API exposes a 500,000-token context window and four reasoning levels, up to xhigh.
• Just 3.3% of risky dual-use prompts passed on SpaceXAI's HackerBench v0.3.

Who should try it: developers and teams that need strong agentic and coding performance at mid-tier pricing, especially if cost per task matters more than raw leaderboard dominance.

Who should skip it: workloads where Terminal-Bench-class performance or peak DeepSWE results are critical may still be better served by Fable 5.1 Max or GPT-5.6 Sol Max, budget differences notwithstanding.