What Is Grok 4.6? SpaceXAI's Frontier Model Explained
xAI, now operating as SpaceXAI, shipped Grok 4.6 just 35 days after Grok 4.5 — and the pitch is unusual for a flagship release: same price, meaningfully higher intelligence. Here's what actually changed, what it costs, where it beats rivals, and the pricing detail that catches a lot of people off guard.
Grok 4.6 gained five points on the Artificial Analysis Intelligence Index in just over a month, at unchanged pricing.
Grok 4.6 Is a Post-Training Refinement, Not a New Base Model
Grok 4.6 is SpaceXAI's newest flagship model, released August 12, 2026 — just five weeks after Grok 4.5. It's built on the same roughly 1.5-trillion-parameter foundation as its predecessor rather than a larger model trained from scratch. The gains come almost entirely from post-training: updated supervised fine-tuning and reinforcement learning across coding and agentic environments, including training data sourced through xAI's Cursor acquisition.
That's worth knowing before you read the benchmark headlines, because "matches GPT-5.6 Sol" is true on the composite Intelligence Index score, but the full picture is more nuanced. Counting individual benchmark wins rather than the composite average, Claude Fable 5 actually takes the most first-place finishes — Grok 4.6's real edge is elsewhere.
What Grok 4.6 Actually Costs
| Tier | Input (per 1M tokens) | Output (per 1M tokens) |
|---|---|---|
| Standard (under 200K prompt tokens) | $2.00 | $6.00 |
| Above 200K prompt tokens | $4.00 | $12.00 |
| Fast variant | $4.00 | $12.00 |
How It Actually Compares
Intelligence Index: tied with GPT-5.6 Sol
Both score 61 on Artificial Analysis's composite measure — one point behind Claude Fable 5 (62) and two behind Claude Opus 5 (63).
Strong on agentic and knowledge work
Grok 4.6 posted a GDPval-AA v2 Elo of 1753 — behind only Claude Opus 5 — and led on AA-Briefcase, a knowledge-work benchmark.
Weaker on hard coding benchmarks
DeepSWE v1.1 lands at 65.9%, behind GPT-5.6 Sol Max's 73%. Terminal-Bench v3.0 sits at 26%, last among the four leading models compared.
Best price-to-intelligence ratio at the frontier
At roughly 60% cheaper than GPT-5.6 Sol or Claude Opus 5 at list price, Grok 4.6's real selling point is what it gets you per dollar, not outright leadership.
Independent testing also found Grok 4.6 resolves tasks in roughly half the turns Opus 5 needs on comparable agent work — a meaningful efficiency edge even where its raw benchmark scores trail. On the flip side, at least one review flagged a higher rate of confident, incorrect answers than its benchmark parity with GPT-5.6 Sol might suggest, so it's worth double-checking factual claims on anything high-stakes.
Where Grok 4.6 Is Actually Meant to Be Used
SpaceXAI positioned Grok 4.6 specifically around long-running agents: researching an unfamiliar topic across many steps, working through a large existing codebase, or turning a rough product idea into a functioning first draft of an app. That's a different pitch than a pure coding-leaderboard release, and it shows in where the model actually wins — knowledge work and legal-reasoning benchmarks, not the coding rows that get the most attention in comparison charts.
Bottom Line
Grok 4.6 doesn't beat Claude Opus 5 or Claude Fable 5 outright, but it ties GPT-5.6 Sol on overall intelligence at roughly 60% of the price — making it the strongest price-to-intelligence option currently at the frontier. It's built for long-running agent work and knowledge tasks rather than pure coding leaderboards, and the 200,000-token pricing cliff is the one detail worth checking before you send a long request.
More AI Guides, Written Clearly
Browse our full library of AI tool guides, updated regularly as models and features change.
Common Questions
What is Grok 4.6?
Grok 4.6 is SpaceXAI's (formerly xAI's) flagship language model, released August 12, 2026. It's a post-training refinement of the same base model as Grok 4.5, tuned for long-running agents, coding, and knowledge work.
How much does Grok 4.6 cost?
Standard pricing is $2 per million input tokens and $6 per million output tokens for prompts under 200,000 tokens. Above that threshold, pricing doubles to $4/$12 per million tokens for the entire request, not just the excess.
Is Grok 4.6 better than GPT-5.6 or Claude Opus 5?
It ties GPT-5.6 Sol on the Artificial Analysis Intelligence Index and trails Claude Opus 5 and Claude Fable 5 by one to two points. Its real advantage is price: it's roughly 60% cheaper than either at list rates.
Where can I use Grok 4.6?
It's available now in Cursor, Grok Build, the xAI API (as grok-4.6), and through partners including OpenRouter, Vercel, and Cloudflare.
More from the creator
Explore two other projects from the same creator.