Grok 4.6 Pricing: API Costs and Plans (August 2026)

Two very different questions hide behind the phrase "Grok 4.6 price". One is about the API — what a developer pays per token. The other is about the consumer subscription — what SuperGrok costs per month. They are separate products with separate bills, and the numbers are nowhere near each other.

This page answers both, with every figure taken from xAI's own pricing and rate-limit documentation on 23 August 2026. Grok 4.6 shipped on 12 August 2026, so most pricing content written before that date describes Grok 4.5.

What does Grok 4.6 cost on the xAI API?

Here is xAI's full published text-model price list. All figures are USD per 1 million tokens, verified against docs.x.ai/developers/pricing on 23 August 2026.

Three things about this table matter more than the headline rate.

The 200K threshold is a cliff, not a slope. xAI's wording is explicit: "requests whose prompt reaches the listed token threshold are billed at the higher rate for all tokens in the request." A 205K-token prompt is billed at $4/$12 on all 205K tokens — not just the 5K above the line. Sending 8% more context can cost you 100% more money.

Cached input on 4.6 is 67% more expensive than on 4.5 — $0.50 versus $0.30. Cache reads are the token class that agent loops hit hardest, because every turn re-sends the accumulated conversation. If you migrated a long-running agent from 4.5 to 4.6 and your bill jumped more than you expected, this is usually why.

The "faster version at twice the price" is a flag, not a model. xAI's launch post mentions a fast variant at double the cost. There is no grok-4.6-fast model ID. It is Priority Processing — set service_tier: "priority" on a Chat Completions or Responses request and you get higher scheduling priority at a flat 2× multiplier on every token type, including cached and reasoning tokens. You are only billed the premium when the response confirms "service_tier": "priority".

Are you looking at API pricing or a Grok subscription?

If you want to chat with Grok in an app, none of the numbers above apply to you. Consumer access is sold as a flat monthly subscription with usage caps, not per token. These are the in-app prices listed for the Grok app (version 1.4.28) as of 23 August 2026.

There is a second route: X Premium bundles Grok with the X platform. On iOS those tiers list at $4/month for Premium Basic, $11/month for Premium, and $50/month for Premium Plus (or $42, $114.99 and $490 annually). App-store prices carry Apple's platform markup — buying on the web is cheaper, so check xAI's own plan page before subscribing through a phone.

The practical rule: a subscription is a fixed monthly ceiling with rate limits; the API is uncapped and metered. If you are building software, you want the API. If you are a heavy individual user, the $30 tier beats API pricing for conversational use almost every time.

Does Grok charge extra for live search and tool use?

Yes, and this is the single most common surprise on an xAI invoice. Server-side tools bill per call, on top of the tokens those calls generate.

Two mitigations are worth knowing. First, only successful tool executions are billed — failed calls, invalid parameters and upstream errors are not charged. Second, search results land in your context as input tokens, so the per-call fee is usually the smaller half of what a search-enabled agent actually costs.

The x_search tool is the one genuinely non-substitutable thing xAI sells: licensed access to the X firehose at $5 per 1,000 calls. No competitor offers it.

What does a realistic Grok 4.6 workload actually cost?

1. Support assistant, 100,000 requests/month. Each request: 8,000 input tokens of which 6,000 are a cached system prompt and knowledge base, plus 700 output tokens.

• Cached input: 600M × $0.50 = $300
• Fresh input: 200M × $2.00 = $400
• Output: 70M × $6.00 = $420
• Total: $1,120/month. Without caching the same job costs $2,020 — caching saves $900, or 45%.

2. Repo-wide coding agent, one run. 240,000 input tokens (180,000 cached) and 30,000 output tokens. The prompt crosses 200K, so the whole request bills at long-context rates: 60,000 × $4 + 180,000 × $1 + 30,000 × $12 = $0.78 per run. Trim the same job to a 190,000-token prompt (140,000 cached) and it costs $0.10 + $0.07 + $0.18 = $0.35. Twenty-six percent more context, 123% more cost.

3. Research agent with live search, 10,000 queries/month. Three web searches per query, 25,000 input tokens and 1,200 output tokens each.

• Input: 250M × $2.00 = $500
• Output: 12M × $6.00 = $72
• Search: 30,000 calls × $5/1,000 = $150
• Total: $722/month — with tool fees at 21% of the bill.

Is Grok 4.6 cheaper than Claude Opus 5, GPT-5.6 and Gemini?

All figures below are USD per 1M tokens, taken from each vendor's own pricing documentation on 23 August 2026, at each model's standard short-context tier.

Grok 4.6's standout number is output. At $6/1M it undercuts Claude Opus 5 by 4.2×, GPT-5.6 Sol by 3.3×, Kimi K3 by 2.5× and Gemini 3.1 Pro by 2×, while scoring 61 on the Artificial Analysis Intelligence Index — sixth of 186 models measured, and level with GPT-5.6 Sol. Artificial Analysis puts it "60%+ below Claude Opus 5 and GPT-5.6 Sol" on headline rates.

But per-token price is not cost per job. Artificial Analysis measured Grok 4.6 at $0.8367 per completed task across its evaluation set — the same figure as Kimi K3, which is 2.5× more expensive per token. The reason is verbosity: running the Intelligence Index consumed 72M output tokens and $1,068 in total spend. Against DeepSeek V4-Pro the gap is wider still — DeepSeek is roughly an order of magnitude cheaper per task for a single-digit deficit on the index.

The honest positioning: Grok 4.6 is the cheapest frontier-scoring model for output-heavy work, but it is not the cheapest way to complete a task. If your workload is well-scoped and a mid-tier model can do it, a mid-tier model will be cheaper. For deeper capability context, see our breakdown of the Grok 4.6 benchmarks.

How do you reduce your Grok 4.6 bill?

1. Stay under 200K tokens per prompt. This is the highest-leverage change available. Retrieve the files an agent needs instead of dumping a repository; the cliff doubles everything.
2. Structure prompts for cache hits. Cached input is $0.50 against $2.00 — a 75% discount. Keep the stable part of your prompt (system instructions, schemas, documentation) at the front and unchanged between calls.
3. Drop to grok-4.3 or grok-build-0.1 where you can. Grok 4.3 costs $1.25/$2.50 with a 1M context and no 200K cliff until 200K; grok-build-0.1 is $1.00/$2.00 and purpose-built for coding agents. Neither needs 4.6's reasoning depth for classification, extraction or routine edits.
4. Turn live search off when you do not need it. An agent that searches on every turn adds $5 per 1,000 turns before a single token is counted, and drags search results into context as billable input.
5. Tune reasoning effort. Grok 4.6 supports low, medium, high (the default) and xhigh. Reasoning tokens bill as output at $6/1M, so leaving everything on high is a real line item.
6. Skip Priority Processing unless latency is the product. The 2× multiplier applies to every token type including cache reads.

What are the rate limits and is there a free tier?

xAI assigns rate limits by cumulative API spend since 1 January 2026. Tiers unlock automatically.

Note that Tier 0's limits are already generous — 150 requests per second and 50M tokens per minute — so rate limits are rarely the binding constraint on a new xAI account. Billing is prepaid: you buy credits, and auto top-up has a $25 minimum with a cap of five top-ups per 24 hours. Monthly invoiced billing is disabled by default and requires contacting sales.

FAQ

How much does Grok 4.6 cost?

On the xAI API, $2.00 per million input tokens, $0.50 per million cached input tokens and $6.00 per million output tokens for prompts under 200K tokens. Prompts of 200K or more are billed at $4.00/$1.00/$12.00 on the entire request. Consumer access is separate: Grok app subscriptions run $10 to $300 per month.

Is Grok 4.6 free?

Not on the API. xAI publishes no free credit allowance — you prepay credits, with a $25 minimum auto top-up. Grok's consumer apps do have a free tier with capped message and image limits, and Grok 4.6 access on the free tier is limited. For unmetered developer testing you would need the free tier of a third-party aggregator, not xAI directly.

What is the difference between SuperGrok and the Grok API?

SuperGrok is a consumer subscription: a flat monthly fee ($10 to $300) that buys chat access in Grok's apps with usage caps. The API is metered per token with no monthly fee and no caps, priced at $2/$6 per million. You cannot use a SuperGrok subscription to make API calls, and API credits do not unlock app features.

Does Grok charge extra for live search?

Yes. Web Search, X Search and Code Execution each cost $5 per 1,000 successful calls on top of token charges. File Attachments cost $10 per 1,000 and Collections Search $2.50 per 1,000. Failed tool calls are not billed. Search results also enter your context as billable input tokens, which is usually the larger cost.

Is Grok 4.6 cheaper than Claude Opus 5?

Per token, substantially: $2/$6 against Opus 5's $5/$25, so 2.5× cheaper on input and 4.2× cheaper on output. Per completed task the gap narrows because Grok 4.6 is verbose. Opus 5 also offers a 1M context at flat pricing, while Grok doubles its rate above 200K tokens.

What is Grok 4.6's context window?

500,000 tokens, unchanged from Grok 4.5. It accepts text and image input with text-only output and no output token cap. Note that the pricing tier changes at 200K, so the top 300K of that window costs double — the effective "cheap" window is 200K.

Does Grok Build have separate pricing?

Two routes. The API model grok-build-0.1 costs $1.00 input, $0.20 cached and $2.00 output per million tokens with a 256K context — the cheapest model xAI publishes. The Grok Build agent product is bundled into SuperGrok and X Premium Plus subscriptions, with the deeper agentic features gated to the higher tiers.

Why did my Grok bill go up after moving from 4.5 to 4.6?

Two reasons, and neither is the headline rate. Cached input rose from $0.30 to $0.50 per million — a 67% increase on exactly the token class that agent loops consume most. And Grok 4.6 is measurably more verbose than 4.5, so the same work generates more billable output tokens even though the $6 output rate did not move.

The decision rule

Use Grok 4.6 when you need frontier-level reasoning and your workload is output-heavy — code generation, long-form synthesis, agentic loops that write a lot. Its $6 output rate is the cheapest in that capability band by a clear margin.

Do not use it as a general cost-minimiser. If your prompts routinely exceed 200K tokens, Gemini 3.1 Pro or Claude Opus 5 have friendlier long-context economics. If the task is routine, grok-4.3, grok-build-0.1 or DeepSeek V4-Flash will finish it for a fraction of the price. And if you are an individual rather than a developer, the $30 SuperGrok subscription beats metered API access for conversational use nearly every time.

For the wider picture on what changed in this release, read our Grok 4.6 launch guide and the Grok 4.6 vs Grok 4.5 comparison. If you are evaluating xAI's developer stack rather than just the model, our guide to Grok Build, Skills and Connectors covers the tooling layer these prices sit under.

Prices verified against xAI's published documentation on 23 August 2026. xAI has changed Grok pricing several times in the past year — confirm on docs.x.ai before committing to a budget.