What is Grok 4.6?
Grok 4.6 is a closed reasoning model from xAI, which now brands itself SpaceXAI. It builds on Grok 4.5 with a longer training run aimed at agents that stay on a task for many steps, and xAI says it tests and checks its own work more often. It reads text and images, writes text, and works across a 500K-token context with four reasoning effort levels.
It is still served on the xAI API and in Cursor, and xAI also launched it on Grok Build, OpenRouter, Vercel and Cloudflare. But xAI and Cursor now both recommend Grok 4.7, which costs the same and scores higher on every test in xAI's own launch table.
Is Grok 4.6 still available now that Grok 4.7 is out?
Yes. xAI still serves grok-4.6 on its API at the same price, but it now points code and chat work to Grok 4.7.
- Price and speed
- Grok 4.7 costs the same $2 / $6 per million tokens and, xAI says, runs at the same speed.
- Base model
- Grok 4.7 uses a new, larger base model and a longer training run on multi-hour tasks.
- Coding scores
- CursorBench 4.0 rises from 40.4% to 46.3%, and Terminal-Bench 4.0 from 20.3% to 37.6%.
- Office work scores
- GDPval rises from 1605 to 1695 Elo, and AA Briefcase v1.1 from 1546 to 1657.
Who is Grok 4.6 best for?
Grok 4.6 is best for long-running agents, first versions of apps, knowledge work on a budget, and teams whose setups are already tuned to grok-4.6.
Long-running agents
xAI tuned it to stay on complex tasks across many steps, and saw it test and check its own work more often on long runs.
First versions of apps
Given a product idea, it researches the domain, structures the app, builds the core interactions and keeps refining through rounds of feedback.
Knowledge work on a budget
xAI's own table puts it at 1753 on GDPVal-AA v2, ahead of Fable 5 at 1741 and GPT-5.6 Sol at 1728, for $2 per million input tokens.
Setups already tuned to grok-4.6
xAI still serves it at the same price as Grok 4.7, so tested prompts keep working while you check the newer model.
What's included with Grok 4.6?
Four effort levels, function calling and structured outputs, image input, xAI's built-in tools, priority processing and a US-only endpoint.
- Four effort levels
- Reasoning effort runs low, medium, high or xhigh, with high as the default.
- Function calling and structured outputs
- It can call your own tools and return answers in a fixed format for agent pipelines.
- Image input
- It reads JPG and PNG images up to 20MiB each, with no limit on how many you send.
- Built-in xAI tools
- Web search and code execution run on xAI's servers for $5 per 1,000 calls, and X search for $5 per 1,000 posts fetched, on top of tokens.
- Priority processing
- Requests can jump the queue for lower latency at twice the standard token rates.
- US regional endpoint
- A US-only endpoint keeps inference in the United States for 10% more; only Grok 4.7 and Grok 4.6 run there.
How much does Grok 4.6 cost?
Grok 4.6 costs $2 per million input tokens and $6 per million output tokens on the xAI API, rising to $4 and $12 once a prompt reaches 200K tokens.
- Cached input $0.50 per million tokens
- 500K-token context, text and image input
- Reasoning effort from low to xhigh, high by default
- Cached input $1.00 per million tokens
- The whole request bills at this rate
Grok 4.6 is reached through the xAI API as grok-4.6, and also runs in Cursor, Grok Build and partners such as OpenRouter, Vercel and Cloudflare. It does not take Batch API requests. Priority processing costs twice the standard rates, and the US regional endpoint adds 10%. Server-side tools cost extra, such as $5 per 1,000 web searches or code runs. Default limits are 150 requests a second and 50 million tokens a minute. In Cursor, a Fast variant costs $4 / $12 per million tokens. xAI's newer Grok 4.7 has the same rates. Full pricing on docs.x.ai. We checked these prices on .
How do readers rate Grok 4.6?
No one has rated Grok 4.6 yet. Be the first.
Sign in to rate. One rating per account.
Grok 4.6 pros and cons
It is cheap for its test scores and developers like it for coding, but Grok 4.7 now does more at the same price and Grok 4.6 can burn a lot of tokens.
Pros
- At $2 / $6 per million tokens it costs a fifth of GPT-6 Astra's input price, and developers on Reddit and Hacker News keep calling it very good at coding for the money.
- Leads xAI's own table on GDPVal-AA v2 knowledge work (1753, against 1741 for Fable 5 and 1728 for GPT-5.6 Sol) and on Harvey LAB legal work.
- High default limits of 150 requests a second and 50 million tokens a minute suit busy agent pipelines.
Cons
- Grok 4.7 costs the same, runs at the same speed and beats it on every test in xAI's launch table, including Terminal-Bench 4.0 at 37.6% against 20.3%.
- Several developers found it burned far more tokens than Grok 4.5 or GPT-5.6 Sol on real tasks and ran slower than Grok 4.5, and some felt it was no smarter, so the low rate does not always mean a cheap job.
- Trails GPT-5.6 Sol and Fable 5 on hard coding tests in xAI's own table: 26% on Terminal-Bench v3.0 against about 34%, and 65.9% on DeepSWE v1.1 against 73% and 70%.
- No Batch API and text-only output, and its 500K context is half of Grok 4.3's; prompts over 200K tokens bill at double rates.
What are the best Grok 4.6 alternatives?
Grok 4.3 for batch jobs and a 1M context, Claude Sonnet 5 for flat-rate long prompts, and Muse Spark 1.3 for a lower token price.
Grok 4.3PaidChoose it when cost and volume matter more than peak quality: it charges $1.25 / $2.50 per million tokens for a 1M-token context and takes Batch API jobs at 20% off, which Grok 4.6 does not.
Claude Sonnet 5FreeChoose it when your prompts are long: it charges a flat $2 / $10 per million tokens across its whole 1M-token context, while Grok 4.6 doubles its rates past 200K tokens.
Muse Spark 1.3PaidChoose it when you want a lower token price and more input types: $1.25 / $4.25 per million tokens on Meta's API, a 1M-token context, and video and PDF input.
Questions about Grok 4.6
What is Grok 4.6?
Grok 4.6 is xAI's model for long-running coding agents and knowledge work, sold on the xAI API at $2 / $6 per million tokens and now followed by Grok 4.7.
Who is Grok 4.6 for?
Grok 4.6 is best for long-running agents, first versions of apps, knowledge work on a budget, and teams whose setups are already tuned to grok-4.6.
Is Grok 4.6 free?
No. Grok 4.6 is a paid tool, from $2.00 / $6.00 per million input / output tokens.
How much does Grok 4.6 cost?
Grok 4.6's paid plans are Prompts under 200K tokens at $2.00 / $6.00 per million input / output tokens and Prompts of 200K tokens or more at $4.00 / $12.00 per million input / output tokens.
What are the best Grok 4.6 alternatives?
The closest alternatives on Recaply are Grok 4.3, Claude Sonnet 5 and Muse Spark 1.3.
Where can I use Grok 4.6?
On the xAI API as grok-4.6, and in Cursor, where its context is 256K tokens and a Fast variant costs $4 / $12 per million tokens. xAI also launched it on Grok Build, OpenRouter, Vercel and Cloudflare, and serves it on a US-only endpoint for 10% more.
Should I use Grok 4.6 or Grok 4.7?
Use Grok 4.7 for new work. xAI calls it its most capable model, charges the same $2 / $6 per million tokens with the same 500K context, and its launch table shows higher scores on every test, such as 46.3% against 40.4% on CursorBench 4.0.
Does Grok 4.6 work with the Batch API?
No. Grok 4.6 does not accept Batch API requests, so every call bills at the standard rate. For large jobs that can wait, Grok 4.3 takes batch requests at 20% off and costs $1.25 per million input tokens to begin with.
What happens when a Grok 4.6 prompt passes 200K tokens?
Once a prompt reaches 200K tokens, xAI bills every token in that request at the long-context rate: $4 per million input tokens, $1 cached and $12 output, double the normal rates. The context window stops at 500K tokens.
Does Grok 4.6 know about recent events?
Not on its own. xAI says its models only know what was in their training data, so for current news you turn on the Web Search or X Search tools, which cost $5 per 1,000 searches or X posts on top of token charges.
Our verdict
Grok 4.6 was good value, but new projects should start on Grok 4.7 instead.
xAI charges $2 / $6 per million tokens for both and serves them at the same speed, and Grok 4.7 scores higher on every test in xAI's launch table.
- Skip it for
- New projects, and batch jobs, which Grok 4.6 does not accept.
- Check first
- If you keep it, watch the tokens each task uses, since some developers saw it burn far more than Grok 4.5.
Grok 4.7
Grok 4.5
Grok 4.20
Ring-2.6-1T
SubQ
Nemotron 3