Grok 4.6 API Pricing Is Here: A Frontier Coding Model at a 60% Discount

SpaceXAI is making a serious push at the frontier price war, and it just got harder to ignore. On August 12 the company released Grok 4.6, a model it says matches GPT-5.6 Sol on the Artificial Analysis Intelligence Index. Two days later, SpaceX closed its $60 billion acquisition of Cursor, the agentic coding tool so many heavy AI users already live in. The two events are connected: Grok 4.6 ships inside Cursor, Grok Build, the SpaceXAI API, OpenRouter, Vercel, and Cloudflare, at API prices that undercut the US frontier by roughly 60%. Here is the exact grok api pricing, how it compares, and what it means for your monthly coding bill.

The new Grok 4.6 API pricing, exactly

Grok 4.6 is priced aggressively for what it claims to deliver. The standard tier costs $2 per million input tokens and $6 per million output tokens. A faster variant costs exactly twice that, $4 per million input and $12 per million output, for workloads where latency matters more than the bill.

To put the grok api pricing in context, look at where the frontier sits today. OpenAI’s GPT-5.6 Sol costs $4 per million input and $20 per million output after last week’s cut. Anthropic’s Opus 5, the Fable 5 class flagship, is around $5 per million input and $25 per million output. On the number that dominates agentic coding, output tokens, Grok 4.6 at $6 is 70% cheaper than Sol and 76% cheaper than Opus 5. The 60% headline discount is actually the conservative way to read it.

Frontier intelligence at a discount

The performance claim is the important part. SpaceXAI says Grok 4.6 scores 61 on the Artificial Analysis Intelligence Index, a composite of nine benchmarks. That matches GPT-5.6 Sol at its maximum reasoning level and trails the Fable 5 Max class (Opus 5) by a single point at 62. On agentic coding specifically, Grok 4.6 shows gains over its predecessor across CursorBench, FrontierCode, APEX-Agents, and Terminal-Bench.

That is a different positioning from the cheap-Chinese-model wave. Grok 4.6 is not a budget workhorse that trades capability for price. It is a frontier-adjacent model, within one benchmark point of the top tier, sold at less than a quarter of the output price. For heavy AI users that is the most interesting corner of the current price war: a model you can route serious agentic work to today, at a price that changes the cost equation of a monthly coding stack.

SpaceXAI says the gain came from a longer supplemental training run, curated engineering data, and expanded reinforcement learning on coding and agentic environments. Whatever the recipe, the benchmark position is the meaningful part for cost planning. If Grok 4.6 genuinely sits one point off Opus 5, then a heavy user paying $25 per million output for Opus 5 has a real alternative at $6.

Why Cursor matters for this pricing story

Grok 4.6 availability inside Cursor is the second half of the story, and it is why this is not just another API price cut. On August 14, the same week Grok 4.6 shipped, SpaceX completed its $60 billion acquisition of Cursor. The model is a first-class option in Cursor today, and SpaceXAI is offering 2x included Grok 4.6 usage inside Cursor and Grok Build for the first week so you can trial it without burning your subscription quota.

For heavy AI users this changes the default option set inside the tool they already open every day. If you pay for Cursor and your included usage resets weekly, the 2x Grok 4.6 window is effectively free extra capacity on a frontier-tier model for the first week. After it ends, you keep a direct API path at $2/$6 that is cheaper than keeping the same workload on a more expensive provider routed through the same harness.

The acquisition also signals this pricing is not a short-lived promotion. A $60 billion deal aligns Cursor’s monetization with Grok model economics, which is exactly the setup where keeping model prices low becomes a durable strategy rather than a launch gimmick. Cursor has been a launchpad for model adoption before, and Grok’s placement inside it buys SpaceXAI distribution for its cheapest frontier option.

What this saves a heavy user

The real math depends on your output share, and agentic coding is output-heavy. Consider a coding agent that reads a codebase, plans, and generates large patches, a profile where 60 to 70% of tokens are output. On such a workload the diff between Grok 4.6 at $2/$6 and GPT-5.6 Sol at $4/$20 is large. Over a month of $2,000 in Sol spend at that mix, moving the same workload to Grok 4.6 standard would land closer to $700, before accounting for any quality gaps on hard tasks.

The faster Grok 4.6 tier is the more honest comparison for latency-sensitive agent loops. At $4/$12 it still undercuts Sol on output by 40% and Opus 5 by more than half, while buying the lower-latency path SpaceXAI targets for production agents. For a team that routes most volume to a cheap tier and escalates only the hardest tasks to a premium model, Grok 4.6 standard plus a premium fallback is a clean two-tier routing setup.

Practical moves for your routing stack

First, trial Grok 4.6 inside Cursor during the 2x window. Pick one real workload, run it on Grok 4.6, and record the per-task cost and the output quality on your specific codebase. A benchmark index number is a starting point, not a substitute for measuring on the work you actually do.

Second, separate the models by tier. Use Grok 4.6 standard for high-volume, well-scoped tasks and route truly novel or safety-critical work to a premium model. That two-tier split is where the grok api pricing earns its keep, because you get the 70% output discount on the bulk of your tokens without betting the hardest tasks on it.

Third, watch the reliability question. A young frontier model in a freshly integrated acquisition deserves a burn-in period: Grok 4.6 can change behavior between versions, and its ecosystem is younger than Anthropic’s or OpenAI’s. Keep a fallback provider key ready so your agent harness can route around failures. The cost savings only count if the model is dependable enough to run unattended.

The bottom line on Grok 4.6 pricing

Grok 4.6 is the strongest discount-to-frontier signal since the current price war began. At $2 per million input and $6 per million output, a model that matches GPT-5.6 Sol on the Artificial Analysis Intelligence Index is priced 60 to 75% below the US frontier on output tokens. It ships inside Cursor on the back of a closed $60 billion acquisition, which makes the low pricing look like a durable strategy, not a launch stunt.

For heavy AI users the play is clear. Bank the 2x Cursor window to trial Grok 4.6 on real work, build a two-tier routing split that sends high-volume tasks to the $2/$6 tier, and measure per-task cost before you commit a critical workload to it. The price war just added a serious new player at the coding harness level, and the cheapest frontier-tier option on the board is now a Grok.