---
title: "Grok 4.6 Lands: What Is New, What It Costs, and the Catch"
description: "xAI shipped Grok 4.6 on August 12, 2026, an upgrade to Grok 4.5 built for long-running agents and heavier interactive work. It keeps the same 500,000 token context and the same $2 and $6 headline API..."
url: https://www.brandligo.com/blog/grok-4-6-launch/
date: 2026-08-14
modified: 2026-08-14
author: "Brandligo Editor"
image: https://www.brandligo.com/blog/wp-content/uploads/2026/08/Grok-4.6-AI-coding-model.jpg
categories: ["AI News"]
type: post
lang: en
---

# Grok 4.6 Lands: What Is New, What It Costs, and the Catch

xAI shipped Grok 4.6 on August 12, 2026, an upgrade to Grok 4.5 built for long-running agents and heavier interactive work. It keeps the same 500,000 token context and the same $2 and $6 headline API price, and it adds a new top-tier reasoning mode called xhigh.

So this is a refinement, not a reinvention. xAI’s own benchmark numbers put it at rough parity with the best models rather than out in front, and the most interesting detail for anyone building on it is a pricing rule that can quietly double the bill.

Here is what actually changed, what it costs, and who should care.

## What is Grok 4.6?

Grok 4.6 is xAI’s newest Grok model, released on August 12, 2026, taking text and image input and producing text output with no set output limit. xAI says it builds on Grok 4.5 with a focus on long-running agents and more ambitious interactive and visual work, and that it holds up better across the many steps of a complex project, testing and checking its own output more as a task drags on. The model keeps a 500,000 token context window, which is roomy but, as several analysts note, still the smallest of the current frontier models. One thing about the name: xAI now sits inside Elon Musk’s SpaceXAI, the brand its own website and developer docs carry, though the model and the API still go by xAI and Grok.

## What does Grok 4.6 cost to use?

Grok 4.6 costs $2 per million input tokens and $6 per million output tokens through the API, as long as your prompt stays under 200,000 tokens. Cached input runs $0.50 per million on that tier. Cross the 200,000 token line and the rates step up to $4 for input, $12 for output, and $1 for cached input. There is also a faster serving variant that costs double the standard rate. Those figures come straight from xAI’s release notes, and they are the same headline rates the previous version carried, according to launch-day analysis.

| Grok 4.6 API rate (per 1M tokens) | Prompt under 200K | Prompt 200K and over |
| --- | --- | --- |
| Input | $2 | $4 |
| Cached input | $0.50 | $1 |
| Output | $6 | $12 |

## The 200K token pricing cliff is the part to watch

Once a prompt reaches 200,000 tokens, Grok 4.6 reprices the whole request at the higher long-context rate, not just the tokens above the line. That is the catch buried in the pricing, and independent analysis of the launch flagged it directly. In practice it means an agent that stuffs a big codebase or a long chat history into a single call can flip from $2 and $6 to $4 and $12 for the entire request, doubling the cost of that turn. If your workload lives near that boundary, a small change in how much context you pack per call can swing your bill more than the model choice itself. Trimming prompts to stay under 200,000 tokens, or splitting work across calls, is worth real money here.

## What actually changed from Grok 4.5?

The headline change is a new reasoning-effort level called xhigh, which sits above the existing low, medium and high settings and lets the model think harder when you ask it to. Beyond that, xAI’s pitch is stamina: better performance on long, multi-step agent runs, and more self-checking as a job stretches out. The context window and the headline $2 and $6 rates match Grok 4.5, so on paper the sticker price did not move. What did move, per launch analysis, is cached input, which rose from around $0.30 to $0.50 per million. That sounds tiny, but agent loops reuse cached context constantly, so a two-thirds jump there is a genuine cost story for anyone running Grok at scale.

## How good is Grok 4.6? Here is what the benchmarks say

xAI’s own numbers put Grok 4.6 at an AA Intelligence Index score of 61, which lands it near the top of the pack without a clear lead over it. The rest of the scorecard leans on agent and coding tasks, which is where xAI is aiming this release. Treat these as vendor-reported figures, measured by the maker on its own setup, so they are a starting point rather than the last word. Here is the official scorecard.

| Benchmark | Grok 4.6 (xAI-reported) |
| --- | --- |
| AA Intelligence Index | 61 |
| GDPVal-AA v2 | 1753 |
| CursorBench v3.2 | 69.9% |
| DeepSWE v1.1 | 65.9% |
| FrontierCode v1.1 | 61.3% |
| APEX-Agents | 57.5% |
| Terminal-Bench v3.0 | 26% |

## Where can you use Grok 4.6?

You can use Grok 4.6 today through the xAI API console, Cursor, Grok Build, OpenRouter, Vercel and Cloudflare. That spread means most developers can reach it inside tools they already use, without moving their whole stack. xAI is also running a launch sweetener: double the included usage for the first week inside Grok Build and Cursor, which is a cheap way to test the model on real work before you commit budget to it. The model ID for the API is grok-4.6, so pointing an existing xAI integration at it is a one-line change in most setups.

## How does Grok 4.6 compare to ChatGPT, Gemini and Claude?

Grok 4.6 competes head-on with the frontier models from [OpenAI](https://www.brandligo.com/companies/openai), [Google](https://www.brandligo.com/companies/google) and Anthropic, and its pitch is strong agent and coding work at a lower sticker price rather than a clear quality crown. On raw intelligence scores it is in the same neighbourhood as the top [ChatGPT](https://www.brandligo.com/brands/chatgpt) and [Gemini](https://www.brandligo.com/brands/google-gemini) models, not obviously ahead. Where it tries to win is cost per token on long agent runs. If you want the full picture on what the big assistants charge, our guide to [what ChatGPT, Gemini and Claude cost in 2026](https://www.brandligo.com/blog/ai-assistant-pricing-2026/) breaks it down, and you can see how the newest rivals stack up in our writeups on [Claude Opus 5](https://www.brandligo.com/blog/claude-opus-5-launch/) and [Meta’s laptop-sized Muse Glimmer](https://www.brandligo.com/blog/meta-muse-glimmer/). The short version: Grok 4.6 is a price-and-agents play, not a knockout blow.

## Who should use Grok 4.6, and who should skip it?

Grok 4.6 makes the most sense if you are building long-running coding or agent workflows and want frontier-level results without paying frontier-level prices. The new xhigh mode and the better multi-step stamina are aimed squarely at that crowd. It is a weaker pick if you routinely feed models enormous prompts, since the 500,000 token window is the smallest among the current frontier models and the price cliff punishes very large requests. It also is not the automatic choice if you need a proven quality lead for hard reasoning, because the scores show parity, not dominance. For most teams already on Grok, though, the upgrade is low-risk. Same price, same context, a stronger reasoning gear, and more endurance on the jobs that actually eat tokens.

## The takeaway

If you already run Grok 4.5, this upgrade is easy to justify: same headline price, same context, a stronger reasoning mode, and better stamina on long jobs. Just keep an eye on that 200,000 token line, because a handful of oversized prompts can double what you pay without changing a line of your code. Point an integration at grok-4.6, run it on your real workload for a week while the promotional usage lasts, then read the bill before you switch anything over for good.

## Quick questions

**When did Grok 4.6 come out?** xAI released it on August 12, 2026.

**How much does Grok 4.6 cost?** $2 per million input tokens and $6 per million output tokens for prompts under 200,000 tokens, rising to $4 and $12 at or above that size, with cached input at $0.50 and $1.

**What is new compared with Grok 4.5?** A new xhigh reasoning level, better performance on long agent runs, and more self-checking, while the context window and headline price stay the same.

**What is Grok 4.6’s context window?** 500,000 tokens, which xAI has kept unchanged from Grok 4.5.

**About this article.** Written and fact-checked by the Brandligo editorial desk. AI tooling was used to gather and cross-check sources; every figure here was verified against the primary sources before publication, including xAI’s official [Grok 4.6 announcement](https://x.ai/news/grok-4-6) and its [developer release notes](https://docs.x.ai/developers/release-notes), with launch-day pricing analysis cross-checked against [DigitalApplied](https://www.digitalapplied.com/blog/grok-4-6-launch-pricing-agentic-benchmarks-2026). Published August 14, 2026. If you spot something out of date, tell us at info@brandligo.com.