Two weeks ago, if you'd asked which AI models sat at the very top, the answer was a two-horse race between Anthropic and OpenAI. On August 12, 2026, xAI — now styled SpaceXAI — crashed that party with Grok 4.6, and the interesting part isn't just that it's good. It's that it matches OpenAI's flagship on an independent benchmark while costing a fraction of the price, and it landed inside the tools developers actually use on day one. For anyone who assumed Grok was a step behind, this is the release that closes the gap — and it does it on price, which is the part that should get your attention.
This is the honest breakdown: exactly where Grok 4.6 lands on the leaderboard, what it costs versus the models it now rivals, where it's already available, and the caveats worth keeping in mind before you crown it. The headline is real, but the details are what make it useful.
Quick summary: xAI's Grok 4.6 launched August 12, 2026. On Artificial Analysis's independent Intelligence Index it scores 61 - tying OpenAI's GPT-5.6 Sol for joint-third, behind only Claude Opus 5 (63) and Claude Fable 5 (62), and up 5 points from Grok 4.5. It costs $2 per million input tokens and $6 per million output (a faster variant is double), with a 500K-token context window - far cheaper than Opus 5 ($5/$25) or GPT-5.6 Sol ($5/$30). It shipped into Cursor the same day and GitHub Copilot two days later (Aug 14). On Artificial Analysis's independent GDPval agentic test it ranks #2, behind only Claude Opus 5. Musk says Grok 4.7 is '3-4 weeks' away.
The Number That Matters
Artificial Analysis, the independent firm that scores models on the same battery of tests, put Grok 4.6 at 61 on its Intelligence Index. That's a five-point jump over Grok 4.5 and, more importantly, it ties GPT-5.6 Sol — OpenAI's flagship — for joint-third place overall. Only Anthropic's Claude Opus 5 (63) and Claude Fable 5 (62) sit above it. In plain terms: Grok 4.6 is now genuinely a frontier model, not a value alternative that trails the leaders by a wide margin. It even overtook Kimi K3, the strongest open-weight model, on the index. For a lab that many wrote off as perpetually one step behind, matching OpenAI's best on an independent test is the real story.
The Price Is the Weapon
Here's where it gets uncomfortable for the incumbents. Grok 4.6 costs $2 per million input tokens and $6 per million output. Compare that to the models it now ties or trails only slightly: GPT-5.6 Sol is $5 in and $30 out; Claude Opus 5 is $5 and $25. That means Grok 4.6 delivers frontier-tier intelligence at roughly a fifth of the flagship output price. There's a faster variant at double the cost, very long prompts above 200K tokens are also billed at double, and cached input is cheaper still - but the base pricing is the point: xAI is undercutting the frontier, not the mid-tier. For anyone running real volume through an API, matching the best on quality while paying a fraction on price is exactly the kind of pressure that has been dragging AI costs down all year.
It's Already Where You Code
A model is only as useful as the places you can actually use it, and xAI moved fast here. Grok 4.6 shipped into Cursor the same day it launched (it even leaked a few hours early), along with the xAI API, OpenRouter, Vercel and Cloudflare. GitHub Copilot followed two days later, on August 14, across eight development surfaces including VS Code, Visual Studio, JetBrains and Xcode. That day-one-to-day-two availability inside the tools developers already live in is a big part of why this launch landed harder than a benchmark chart alone would suggest — you could try it immediately, in your editor, against your own code.
| Model | AA Intelligence Index | Price /M (in-out) |
|---|---|---|
| Claude Opus 5 | 63 | $5 / $25 |
| Claude Fable 5 | 62 | $10 / $50 |
| Grok 4.6 | 61 | $2 / $6 |
| GPT-5.6 Sol | 61 | $5 / $30 |
| Grok 4.5 (previous) | 56 | $2 / $6 |
The Caveats Worth Keeping
Three honest notes before you switch everything over. First, some of the flashier gains are xAI's own numbers: the jumps on the DeepSWE coding test (to 65.9) and APEX-Agents (to 57.5) are vendor-reported, not independently reproduced, so treat them as claims. The independent data point that does hold up is strong, though — on Artificial Analysis's own GDPval agentic-work test, Grok 4.6 ranks second, behind only Claude Opus 5. Second, xAI still doesn't disclose the model's parameter count, and it's text-and-image-in, text-out — not a full multimodal model. Third, this is a fast-moving target: Musk says Grok 4.7 is three to four weeks out, so today's leaderboard snapshot will shift again soon. None of that dents the core result; it just means you should verify on your own workload rather than trust the launch chart.
- Grok 4.6 (Aug 12, 2026) scores 61 on Artificial Analysis's independent index - joint-third, tying GPT-5.6 Sol, behind Opus 5 (63) and Fable 5 (62).
- That's +5 over Grok 4.5, and it overtook the top open-weight model, Kimi K3.
- Price is the weapon: $2/$6 per million tokens vs $5/$30 for GPT-5.6 Sol - frontier quality at a fraction of the cost.
- Available in Cursor same-day; GitHub Copilot two days later (Aug 14) across eight surfaces.
- Independent win: #2 on Artificial Analysis's GDPval agentic test, behind only Claude Opus 5.
- Caveats: DeepSWE/APEX gains are vendor-claimed, params undisclosed, and Grok 4.7 is reportedly weeks away.
01Is Grok 4.6 as good as GPT-5.6 or Claude?
On Artificial Analysis's independent Intelligence Index, Grok 4.6 scores 61 - exactly tying GPT-5.6 Sol and sitting just behind Claude Opus 5 (63) and Fable 5 (62). So it's genuinely frontier-tier now, matching OpenAI's flagship and trailing Anthropic's best by only a couple of points.
02How much does Grok 4.6 cost?
$2 per million input tokens and $6 per million output, with a faster variant at double that and cheaper cached input. That's roughly a fifth of GPT-5.6 Sol's output price ($30) and well under Claude Opus 5 ($25), making it the cheapest model at the frontier.
03Can I use Grok 4.6 in Cursor or GitHub Copilot?
Yes to both. It shipped into Cursor on launch day (Aug 12) and arrived in GitHub Copilot on Aug 14 across eight surfaces including VS Code, Visual Studio, JetBrains and Xcode. It's also on the xAI API, OpenRouter, Vercel and Cloudflare.
04Are Grok 4.6's benchmark scores trustworthy?
Mixed. Its Intelligence Index score of 61 and its #2 GDPval agentic ranking are from independent Artificial Analysis testing. But the eye-catching DeepSWE (65.9) and APEX-Agents (57.5) coding figures are xAI's own, not independently reproduced - treat those as vendor claims.
05Should I switch to Grok 4.6?
If you run high volumes and want frontier quality at a fraction of the price, it's compelling - especially for coding, where it's already in Cursor and Copilot. But test it on your own tasks first, and note Grok 4.7 is reportedly only weeks away, so the landscape will shift again soon.
Grok 4.6 is the clearest sign yet that the frontier is now a four-way race, not a two-horse one — and that the price of top-tier AI is collapsing as the field crowds. The only way to know whether it beats your current model on your actual work is to run the same prompts through both. LumiChats puts many leading models — across xAI, OpenAI, Anthropic and more — under one login at a pay-per-day price, so you can pit Grok 4.6 against the flagships head-to-head and pay only for the days you use it.
