Grok 4.6 Pricing, Access, and Usage Limits Explained
Grok 4.6 costs $2 per million input tokens and $6 per million output tokens. Here's where to access it and how the launch-week double usage promo works.
What does Grok 4.6 cost?
Grok 4.6 is priced at $2 per million input tokens and $6 per million output tokens through xAI’s API. There’s also a fast variant that runs at twice the price of the standard model, aimed at people who want lower latency and are willing to pay for it. Compared to frontier models from OpenAI and Anthropic, that pricing is notably lower, with multiple sources describing Grok 4.6 as roughly half the cost (or less) of comparable GPT-5.6 and Claude tiers while landing close to them on benchmark performance.
TL;DR
- Grok 4.6 costs $2 per million input tokens and $6 per million output tokens, with a fast variant priced at double that rate for lower latency.
- The model launched alongside a double usage promotion in Cursor and Grok Build during launch week, meaning the fast variant’s 2x cost is roughly offset by 2x usage allowance.
- You can access Grok 4.6 through Cursor, Grok Build, the xAI API, OpenRouter, Vercel, and Cloudflare, giving developers several entry points depending on their existing workflow.
- Benchmark results across GDPval, Cursor Bench, and the Artificial Analysis Intelligence Index place Grok 4.6 close to or ahead of GPT-5.6 Soul and near Claude Opus/Sonnet-tier models, at a fraction of the per-task cost in some comparisons.
- The pricing story is tied directly to xAI’s acquisition of Cursor, which supplied curated coding and reasoning data that fed Grok 4.6’s supplemental training run.
- Grok Bot, xAI’s always-on agent product, runs on top of Grok 4.6 (or is expected to shortly) and represents a separate access surface aimed at non-developers.
- A follow-up model, Grok 4.7, is expected within three to four weeks of the 4.6 launch, according to statements attributed to Elon Musk, with Grok 5 targeted before year end.
Built like a system. Not vibe-coded.
Remy manages the project — every layer architected, not stitched together at the last second.
How does Grok 4.6 pricing compare to GPT-5.6 and Claude?
At $2 per million input tokens and $6 per million output tokens, Grok 4.6 undercuts both GPT-5.6 Soul and Anthropic’s Claude models by a wide margin, based on comparisons made across the sources covering the release. One creator ran a side-by-side test generating two comparable 3D scenes and found Grok 4.6 cost around 38 cents for a task that cost roughly $2 on Claude Opus 5, a rough five-to-one difference for similar output.
That gap matters because token efficiency, not just sticker price, determines real cost. A model that’s cheaper per token but generates far more output tokens to reach the same answer can end up costing the same or more. On the Artificial Analysis cost-per-task chart, Grok 4.6 high lands in a favorable position: its intelligence index score sits near GPT-5.6 Soul Max and Kimi K3, but its cost per completed task is lower than Soul Max’s and roughly in line with or better than Kimi K3’s. Grok 4.5, by comparison, was cheaper still (around 36 cents per task) but scored lower on intelligence, so the jump to 4.6 traded some cost efficiency for a meaningful capability gain.
Where can you access Grok 4.6?
Grok 4.6 is available through several channels as of launch:
- Cursor, the AI coding IDE that xAI acquired, where Grok 4.6 can be selected as a model for coding tasks and sub-agent work.
- Grok Build, xAI’s own coding-focused product that functions similarly to Cursor and is expected to eventually merge with it.
- The xAI API, for direct programmatic access at the standard per-token pricing.
- OpenRouter, Vercel, and Cloudflare, third-party platforms that route requests to Grok 4.6 alongside other models, useful for developers who want to switch between providers without managing separate API keys.
Separately, Grok Bot is xAI’s always-on agent platform, available as a desktop app on Windows, macOS, and Linux, with Android versions planned. Grok Bot is structured differently from a typical chatbot: users interact with a single “chief of staff” agent that delegates tasks to sub-agents, each running in its own cloud-based virtual machine that stays active continuously rather than only during a chat session. At the time of Grok 4.6’s release, it wasn’t fully confirmed whether Grok Bot was already running on the new model or still using a previous version, though xAI has signaled that 4.6 will power it.
What is the launch-week double usage promotion?
During Grok 4.6’s launch week, xAI offered double usage allowances for users running the model inside Cursor or Grok Build. This effectively offsets the cost of the fast variant, which is priced at twice the standard rate: if you’re getting double the usage credit during the promotional period, the higher per-token cost of the fast variant is roughly balanced out for that window. Outside the promotional period, the fast variant’s 2x pricing applies without the usage offset, so it functions as a straightforward speed-for-cost tradeoff once the promotion ends.
- ✕a coding agent
- ✕no-code
- ✕vibe coding
- ✕a faster Cursor
The one that tells the coding agents what to build.
This kind of launch promotion is common in the AI infrastructure space as a way to drive adoption and usage data quickly after a new model ships. For developers evaluating whether to build on Grok 4.6 long-term, it’s worth testing the fast variant during a promotional window and then re-checking cost math once standard pricing resumes.
Is Grok 4.6 worth using over GPT-5.6 or Claude?
Whether Grok 4.6 is worth choosing depends heavily on the use case. On raw benchmark scores, it performs close to or ahead of GPT-5.6 Soul on tasks like GDPval (which measures performance on real economic and professional tasks) and Harvey Bench (legal use cases), and it made large jumps over its predecessor, Grok 4.5, on tests like Terminal Bench. On DeepSuite, a coding-feel benchmark many developers consider a more honest signal, it ranked third behind GPT-5.6 Soul Max and Claude, suggesting it’s strong but not universally the top choice for coding work.
Design and visual output quality is more mixed. In one head-to-head test generating a styled profile card, GPT-5.6 Soul produced a noticeably cleaner result than Grok 4.6, while Grok 4.6 still outperformed Claude 5 in that specific comparison. This suggests Grok 4.6 is a strong generalist and a genuinely price-competitive option for coding and agentic tasks, without being a clean sweep winner across every category.
For anyone building tools where cost per task matters more than shaving the last few percentage points off a benchmark, Grok 4.6’s pricing makes it a reasonable default, particularly for high-volume coding or agent workloads where token costs compound quickly.
Frequently Asked Questions
How much does Grok 4.6 cost per million tokens?
Grok 4.6 costs $2 per million input tokens and $6 per million output tokens. The fast variant costs twice that rate in exchange for lower latency.
Where can I use Grok 4.6 right now?
Grok 4.6 is available in Cursor, Grok Build, the xAI API, and through third-party platforms including OpenRouter, Vercel, and Cloudflare.
What is the double usage promotion for Grok 4.6?
During launch week, xAI offered double usage allowances in Cursor and Grok Build for models including Grok 4.6, which helps offset the higher cost of the fast variant during that period.
Is Grok 4.6 cheaper than GPT-5.6 or Claude?
Yes, based on comparisons cited around the launch, Grok 4.6 is priced substantially lower than GPT-5.6 Soul and Claude models, and in at least one head-to-head test it completed a comparable task for a small fraction of Claude Opus 5’s cost.
Does Grok Bot use Grok 4.6?
Grok Bot is xAI’s always-on agent platform and is expected to run on Grok 4.6, though at the time of the model’s release it wasn’t fully confirmed which version was live on that platform.