FROM THE EDITORS
xAI released Grok 4.7 on September 21, 2026, calling it the company’s most powerful model yet for coding and knowledge work. The model runs on a new base architecture of 2.1 trillion parameters – a 40% scale-up over Grok 4.6 – and was trained on a longer reinforcement-learning schedule weighted toward multi-hour problem-solving, supplemented with internal engineering data from SpaceX operations including Starlink telemetry and manufacturing logs.
For small and medium-sized businesses in the EU, the headline is straightforward: Grok 4.7 is priced at $2 per million input tokens and $6 per million output tokens – the same as its predecessor and roughly half the cost of comparable frontier models – while delivering substantial gains across coding, professional knowledge work, and safety.
== Where Grok 4.7 fits for an EU SMB ==
Professional knowledge work. Grok 4.7 is specifically tuned for tasks performed by lawyers, nurses, financial analysts, and engineers. It scored 1,695 Elo on GDPval and 1,657 on the AA Briefcase agent benchmark – a 111-point jump over Grok 4.6 – putting it in the same tier as Anthropic’s Fable 5.1 for document creation, presentations, and multi-hour office workflows. For a small law firm drafting contracts or a regional accounting practice producing client reports, this means fewer iterations and more reliable output.
Coding and software engineering. On CursorBench 4.0, Grok 4.7 reached 46.3% – outperforming GPT-5.6 Sol (41.7%) – and hit 71.0% on DeepSWE v1.1 at high effort. A development team of two or three can use it inside Cursor or via the Grok API to handle complex refactoring, debugging, and feature implementation that previously required senior-level bandwidth.
Cybersecurity and compliance. This is where the model stands out for European businesses operating under GDPR and NIS2. Grok 4.7 ships with an entirely new safeguard stack that xAI describes as its strongest yet on refusal accuracy and jailbreak resistance. On HackerBench v0.3, it let through only 3.3% of risky dual-use prompts while rarely blocking legitimate security work, and it topped LatchBio’s biosafety benchmark at 62.4%. For SMBs that need defensible, well-calibrated AI – not just raw capability – this matters.
== Pricing and access ==
Grok 4.7 (xHigh): $2/M input, $6/M output
Grok 4.6 (High): $2/M input, $6/M output
GPT-5.6 Sol (Max): $4/M input, $20/M output
Fable 5.1 (Max): $10/M input, $50/M output
Benchmark highlights (Grok 4.7):
- CursorBench 4.0: 46.3%
- DeepSWE v1.1: 71.0%
- EEBench: 64.0%
- AA Briefcase (Elo): 1,657
- Harvey Legal: 19.6%
- HealthBench Professional: 56.7%
A fast variant is also available at twice the output speed and twice the price, for latency-sensitive use cases. The model is accessible today through Cursor, Grok Build, the Grok API, and third-party cloud platforms – meaning an EU SMB can integrate it via API without infrastructure investment, or try it free at x.ai/build.
== The bottom line ==
Grok 4.7’s value proposition for European SMBs is not that it beats every model on every benchmark – Fable 5.1 leads on several – but that it delivers frontier-tier performance at half the token cost, with a safety stack calibrated for the kind of regulatory environment EU businesses operate in. For a company that needs coding support, professional document generation, and cybersecurity-aware AI without enterprise-scale spending, Grok 4.7 is worth a serious look.
Source: https://x.ai/news/grok-4-7
Leave a Reply