Strengths
- Output costs $6 per million tokens. In xAI's comparison that is against $20 for the older GPT-5.6 Sol and $50 for Claude Fable 5.1. Even the newer GPT-6.1 Sol charges $10 for output.
- A big jump in agentic coding: on the Artificial Analysis index it rose from 47 to 56 and ranks fourth among models tested in their makers' own tools.
- Strong at professional office work: on AA-Briefcase it sits just behind Claude Opus 5 and Claude Fable 5.1, and it scores 1695 on GDPval-AA.
- It makes things up less than its predecessor: its hallucination rate in Artificial Analysis testing is 29%, down from 34% for Grok 4.6.
- In xAI's table it has the best score on the Harvey Legal Agent Benchmark (19.6%) and the EEBench electrical engineering test (64%).
- Besides the xAI API, companies can get it on Microsoft Foundry since October 2026, at the same $2 input and $6 output price.
Weaknesses
- Very verbose: on one Artificial Analysis task it writes about 81K tokens, against about 36K for Grok 4.6. Real costs and waiting times are higher than the price list suggests.
- It lags on long terminal work: xAI reports 37.6% on Terminal-Bench 4.0, against 57.9% for Claude Fable 5.1 at max effort in the same table.
- Its 500K-token context window is half that of Grok 4.3 and several rivals with 1 million.
- No EU processing: it runs only in US regions. xAI only guarantees the processing location through its US endpoint (10% more expensive), and Microsoft Foundry offers no EU data zone either.
- It is slow: one Artificial Analysis task takes it about 7 minutes on average.
Best for
- coding agents
- agents working on professional tasks on their own
- drafting documents and presentations
- legal and engineering analysis
Not ideal for
- fast chat where instant replies matter
- companies that need data processed in the EU
- inputs over 500K tokens
How to use it in Europe
In Europe you can use it in the Grok app and on grok.com in Expert and Heavy modes, and The X Subscription members also get it in Fast mode. Developers have it in Grok Build and Cursor, and companies pay per use through the xAI API or deploy it on Microsoft Foundry. Data is processed in the US. xAI currently offers guaranteed EU processing for none of its models, not even the cheaper Grok 4.3.
What is coming
Rumor, unconfirmed In September 2026, Elon Musk wrote on X that Grok 4.8 has 2.5 trillion parameters and will be a noticeable improvement. xAI has not released it or given a date. When: no date, Musk mentioned the model in September 2026 →
Rumor, unconfirmed Musk keeps hinting that Grok 5 will be a big leap, but xAI has never published a date and Musk's earlier estimates did not hold. Musk has also been talking about Grok 4.8 and 4.9. When: no date, Musk's earlier Q1 2026 target has passed →
Price compared with similar models
Benchmark results
| Benchmark | Score | Source |
|---|---|---|
| Artificial Analysis Intelligence Index (xhigh) | 46 | artificialanalysis.ai ↗, September 2026 |
| Artificial Analysis Coding Agent Index (Grok Build, xhigh) | 56 | artificialanalysis.ai ↗, September 2026 |
| GDPval-AA | 1695 Elo | artificialanalysis.ai ↗, September 2026 |
| DeepSWE v1.1 (high, vendor-reported) | 71.0% | x.ai ↗ (vendor-reported), September 2026 |
| Terminal-Bench 4.0 | 37.6% | x.ai ↗ (vendor-reported), September 2026 |
Benchmarks cover only a narrow slice of ability and vendors often report them at their best settings. Treat them as a rough guide.
More models from this vendor
- Grok 4.3: An older, cheaper Grok from April 2026 with a 1-million-token window. Good for lots of low-cost API calls.
- Grok Build 0.1: A fast Grok coding model and the cheapest Grok on the API, built for agents that write and fix code on their own.
Similar models from other vendors
- Claude Opus 5.5 (Anthropic): Anthropic's main model for demanding work, coding and long documents, and the one Anthropic recommends starting with.
- DeepSeek V4 Pro (DeepSeek): DeepSeek's largest open-weight model for demanding text tasks at a low price. It does not understand images.
- Gemini 4 Carbon (Google): Google's most capable model, announced on 28 October 2026. So far only selected cybersecurity experts have access.
- Gemini 4 Argon (Google): Google's most capable publicly available model, out in October 2026. Built for demanding analytical work where accuracy matters and made-up answers are costly.
- Muse Wonder (Meta): Meta's strongest model for long-running agents and coding, still released as a preview.
- Mistral Large 4 (Mistral AI): Mistral's largest model, still in preview. Built for demanding tasks, long documents and security analysis with data kept in the EU.