Grok 4.6 API Pricing: Full Costs, 200K Context & 30% Savings

Grok 4.6 API pricing starts at $2 per million input tokens, $0.50 per million cached input tokens, and $6 per million output tokens on xAI's standard API. At 200K prompt tokens, xAI's published long-context rates rise to $4/M input and $12/M output. The model supports a 500K-token context window, but using more context can substantially increase the cost of a request.

The real challenge is predicting your bill. Cached tokens still count toward context length, reasoning increases billable output, and agent tools may carry separate charges. A request near the 200K threshold can become much more expensive without producing a noticeably longer answer. For coding assistants and research agents, token efficiency matters as much as the advertised rate.

LinkModel makes Grok 4.6 more affordable with regular token prices 30% below xAI Standard:** $1.40/M input, $0.35/M cached input, and $4.20/M output for short-context requests. You also get pay-as-you-go billing, $0 minimum spend, $0 monthly platform fees, and unified access to multiple AI models. Explore LinkModel to reduce Grok 4.6 token spending without committing to a monthly platform subscription.

Grok 4.6 API Pricing: Official Rates vs LinkModel

Grok 4.6 uses tiered token pricing, with separate rates for uncached input, cached input, and output. At xAI's standard rates, input costs $2/M and output costs $6/M below the 200K prompt threshold. LinkModel lists corresponding prices that are 30% lower, including for long-context requests.

Grok 4.6 Pricing Table

All prices are in USD per 1 million tokens.

The savings are 30% for each listed token category. The prices above compare the same token quantities, rather than assuming that one provider automatically uses fewer tokens.

The short- and long-context rates are published in the xAI pricing documentation and the LinkModel model catalog.

An important distinction: reasoning output uses the output-token rate. It is displayed separately to explain the billing treatment, not because the same tokens are charged twice.

What Counts Toward Your Grok 4.6 API Bill?

A Grok 4.6 request can contain three billable token components:

- Uncached input: new prompt content processed at the regular input rate.

- Cached input: eligible reused prompt content processed at the lower cached-input rate.

- Billable output: generated output, including reasoning tokens.

For a request that stays in the short-context tier, the calculation is:

Request cost = (Uncached input ÷ 1M × input price) + (Cached input ÷ 1M × cached price) + (Billable output ÷ 1M × output price)

For example, 10K uncached input tokens and 2K billable output tokens cost:

At 50,000 identical requests, that is $1,600 on xAI Standard versus $1,120 on LinkModel, saving $480 in model-token charges.

Does Grok 4.6 API Pricing Include a Subscription?

The prices above are usage-based API token rates, not the price of a Grok consumer subscription or a coding editor's monthly plan.

API token billing and subscription usage quotas are different products. A subscription allowance does not translate directly into a fixed number of Grok 4.6 API requests, because token composition, reasoning, context length, and the application provider's rules may differ.

For developers who want usage-based access, LinkModel offers pay-as-you-go billing without a minimum spend or monthly platform fee

Grok 4.6 Long-Context Pricing: Why the 200K Threshold Matters

Grok 4.6 supports up to 500K context tokens, but xAI's long-context pricing starts at 200K prompt tokens. Once the long-context tier applies, the higher rates are charged across the request, not merely on tokens beyond the threshold.

This is one of the most important differences between Grok 4.6's headline token price and the amount developers may actually pay.

Case Study 1: The 199K vs 200K Pricing Jump

Consider three xAI Standard requests, each producing 10K billable output tokens.

At 199K prompt tokens, the request costs $0.458.

At 200K prompt tokens, xAI's published pricing table places the request in the long-context tier, increasing the cost to $0.920.

That is approximately a 100.9% increase in total token cost for 1K additional prompt tokens.

Why does this happen? The pricing tier changes for the entire request, so both the input and output use the higher rates.

LinkModel also publishes short- and long-context rates, but its model page labels the short tier as up to and including 200K, with long-context pricing above 200K. Therefore, the exact 200K boundary should be checked against the provider's current billing terms rather than assumed to be identical.

For an unambiguous cross-provider comparison, the next example uses 190K and 240K prompts, which fall clearly on opposite sides of both published tier boundaries.

How Much Can Context Reduction Save?

Suppose a coding assistant sends 240K prompt tokens and generates 10K billable output tokens.

After removing irrelevant repository files, outdated conversation history, and duplicate logs, the prompt falls to 190K tokens while producing the same output length.

The revised request costs approximately 59.3% less on either provider.

This reduction comes from two effects: fewer input tokens and a lower pricing tier for the entire request.

The practical lesson is to remove irrelevant context, not necessary context. Aggressive trimming that causes incorrect answers and repeated requests can increase the total cost of completing the task.

Do Cached Tokens Count Toward the 200K Limit?

Yes. Cached input is still part of the total prompt length used to determine the applicable context tier.

For example, a prompt with 150K cached input tokens and 60K uncached input tokens contains 210K prompt tokens.

It qualifies for long-context pricing even though most of the prompt was served from cache.

Caching changes the price of eligible input tokens. It does not make those tokens disappear from the context calculation.

Grok 4.6 Cached Input Pricing: How to Cut Repeated Prompt Costs

Cached Grok 4.6 input costs 75% less per token than uncached input within the same pricing tier. At xAI's short-context rates, that means $0.50/M instead of $2/M. LinkModel offers the corresponding prices at $0.35/M and $1.40/M.

Prompt caching is particularly useful for coding agents, multi-turn conversations, customer-support systems, and applications that repeatedly send stable instructions.

How Grok 4.6 Prompt Caching Works

The xAI API can reuse a matching prompt prefix from an earlier request. This means caching is most effective when the beginning of the prompt remains consistent.

To improve cache reuse:

1. Put stable content first. Place system instructions, reusable examples, and reference material before frequently changing user content.

2. Preserve earlier conversation content. Editing or reordering existing messages can break the cached prefix.

3. Reuse a conversation-specific cache key. On the xAI API, the documented prompt cache key helps related requests reach the appropriate cache.

4. Preserve reasoning context where supported. Changing or omitting reusable reasoning content can reduce cache effectiveness.

5. Check actual cached-token usage. On xAI's Responses API, the cached-token field in usage details shows whether input was served from cache.

These are xAI API behaviors described in its prompt caching best practices. When using LinkModel, verify which cache-related parameters and usage fields its current endpoint supports rather than assuming every xAI-specific feature is identical.

Cache hits are not guaranteed. The first request may be a cache miss, and previously cached content may become unavailable.

Case Study 2: 80% Cached Input Across 100,000 Requests

Consider a repeated workflow with:

- 100K total input tokens per request.

- 80K cached input tokens.

- 20K uncached input tokens.

- 10K billable output tokens.

- 100,000 requests.

All requests remain below the long-context threshold.

With caching, LinkModel's modeled cost is:

20K × $1.40/M + 80K × $0.35/M + 10K × $4.20/M = $0.098 per request

Compared with xAI Standard without caching, the modeled monthly bill falls from $26,000 to $9,800, a 62.3% reduction.

That figure combines two independent advantages: cached-input savings and LinkModel's lower published token prices.

It assumes the stated cache-hit distribution is achieved. Actual spending depends on the cache hits and output tokens recorded by the application.

When Is Grok 4.6 Caching Worth It?

Caching usually provides the greatest benefit when large portions of a prompt are repeated.

For example, a support assistant may reuse the same policy documentation across thousands of conversations. A coding agent may repeatedly reference stable project instructions. A document-analysis application may ask multiple questions about the same source material.

Caching provides less value when nearly all input changes between requests.

The right metric is not simply the number of requests. It is how many input tokens are actually billed at the cached rate.

Grok 4.6 Reasoning Token Pricing: Low, Medium, High, and Xhigh

Grok 4.6 offers four reasoning-effort settings: low, medium, high, and xhigh, with high as the default. Reasoning output is billed at the applicable output-token rate, so increasing reasoning usage can raise costs even when the visible response remains short.

According to the official Grok 4.6 model documentation, the model supports configurable reasoning alongside its 500K context window, image understanding, and tool-calling capabilities.

Are Grok 4.6 Reasoning Tokens Charged Separately?

Reasoning tokens are billable output tokens. They use the same output-token price applicable to the request's context tier.

They should not be counted twice.

For cost estimation, add the billable reasoning tokens and other billable output tokens to obtain total output usage, then apply the output rate once.

For example, a request with 2K visible output tokens and 3K reasoning tokens has 5K billable output tokens, assuming those are the usage components reported by the API.

At xAI's short-context rate, 5K output tokens cost $0.030. At LinkModel's published rate, the corresponding charge is $0.021.

How Much Can Reasoning Increase the Bill?

Consider two possible token-usage profiles for the same 20K-token input prompt.

The higher-reasoning example costs more than twice as much, despite producing the same number of visible output tokens.

These figures demonstrate the effect of different token counts. They are not measured performance results for specific reasoning-effort settings.

Which Reasoning Effort Should Developers Choose?

Higher reasoning effort is not automatically more cost-effective.

A configuration that uses fewer tokens but repeatedly fails may cost more per completed task than one that reasons longer and succeeds on the first attempt.

For production evaluation, use:

Cost per successful task = Total cost of all attempts ÷ Number of successfully completed tasks

This captures the effect of retries, reasoning, and task quality without assuming that the cheapest individual response is always the best choice.

Grok 4.6 API Cost Calculator: Per-Request and Monthly Examples

To estimate Grok 4.6 monthly spending, calculate the average cost of each request and multiply it by expected request volume. Separate short-context and long-context requests, and account for cached input, reasoning output, and tool usage where relevant.

Case Study 3: 100,000 Support Requests per Month

Suppose a customer-support application processes 100,000 requests per month.

Each request averages:

- 5K uncached input tokens.

- 1K total billable output tokens.

- No separately billed server-side tools.

Monthly usage equals 500M input tokens and 100M output tokens.

At this workload, LinkModel's 30% lower token rates save $480 each month without requiring any assumed improvement in caching or token efficiency.

Monthly Grok 4.6 API Budget Comparison

For identical billable token quantities at corresponding standard rates:

For a workload equivalent to $10,000 in monthly xAI Standard token charges, LinkModel's published token prices reduce the modeled spending to $7,000.

That is $36,000 in annual savings if token usage and pricing remain unchanged.

These estimates exclude separately billed tools and other charges outside the compared model-token rates.

Why Request Volume Alone Is Not Enough

Two applications processing the same number of requests can have very different bills.

A classification service may send 1K input tokens and receive 100 output tokens. A coding agent may repeatedly send large repository contexts, generate substantial reasoning output, and invoke tools.

For more reliable forecasts, track average tokens per request, cache-hit rate, the percentage of requests reaching long-context pricing, and average tool usage.

The most useful budget is based on your workload's token distribution, not a generic cost per request.

Grok 4.6 Agent Pricing: Web Search, X Search, and Code Execution

Grok 4.6 agent workflows can incur server-side tool charges in addition to model-token costs. This matters for research assistants and coding agents because a short final answer may require multiple searches or code executions.

xAI publishes separate rates for its server-side tools. These charges should not be confused with the price of model-generated tokens.

Official xAI Server-Side Tool Pricing

These are the published xAI direct-API rates, as described in the official xAI pricing documentation.

Tool availability and billing are provider-specific. Do not assume that every xAI-hosted tool or parameter is available through LinkModel, or that LinkModel's 30% model-token discount applies to tool charges.

Case Study 4: A Research Agent With 10 Web Searches

Consider an xAI direct-API research workflow with:

- 20K uncached input tokens.

- 8K billable output tokens.

- 10 billable Web Search calls.

In this example, Web Search accounts for approximately 36.2% of the total cost.

This is why an agent's cost should not be estimated from input and output tokens alone.

For research-heavy workflows, reducing redundant searches may deliver more savings than shortening a small prompt.

Total agent cost = Model-token cost + Applicable server-side tool charges

The example uses xAI's published direct-API rates. It does not assume that the same tool configuration is available through every third-party API provider.

Grok 4.6 Batch API Pricing: Standard vs Batch vs Priority

xAI offers a 20% Batch API token discount for Grok 4.6. Batch requests are processed asynchronously, making them suitable for workloads that do not require immediate responses.

Priority Processing serves a different purpose: higher scheduling priority at 2× standard token rates when the request is actually served at the priority tier.

Grok 4.6 Standard vs Batch Prices

The Batch discount also applies to the corresponding long-context token rates.

For a short-context workload using 100M uncached input tokens and 20M billable output tokens:

LinkModel's standard Grok 4.6 token prices are 12.5% below the equivalent xAI Batch token prices.

However, the services are not operationally identical: Batch is designed for asynchronous processing, while standard requests are appropriate for interactive applications.

When Should You Use Batch or Priority?

Choose Batch when a workload can tolerate delayed completion, such as large-scale document classification or offline processing.

Choose standard processing when users expect immediate interactive responses.

Consider Priority Processing when lower latency has enough business value to justify its higher token rate.

For xAI Priority Processing, the 2× rate applies when the response confirms that it was served at the priority tier. Requests served at the default tier use standard pricing.

xAI also publishes a 10% token-rate premium for its US regional endpoint, where supported. That is an xAI-specific regional option, not a universal surcharge on Grok 4.6 usage through other providers.

Grok 4.6 vs Grok 4.7: Is the Newer Model More Expensive?

Grok 4.6 and Grok 4.7 have the same published xAI Standard token rates, including $2/M input, $0.50/M cached input, and $6/M output below 200K prompt tokens.

Both models also offer a 500K-token context window.

The choice should therefore depend on task quality, reasoning-token usage, latency, and compatibility, rather than headline token price alone.

For an established Grok 4.6 application, migrating to 4.7 should involve regression testing against representative tasks.

A newer model is not automatically cheaper to operate simply because its published price per million tokens is unchanged.

How to Reduce Grok 4.6 API Costs

The most effective way to reduce Grok 4.6 spending is to optimize the entire request lifecycle: provider pricing, context length, cache reuse, reasoning usage, and tool calls.

For a coding assistant near the 200K threshold, context reduction may be the highest-impact optimization.

For a support system with repeated instructions, caching may matter more.

For a research agent, server-side tool usage can represent a substantial share of spending.

There is no single optimization that dominates every workload. The right approach is to measure where the bill comes from and prioritize the largest controllable cost.

Why Choose LinkModel for Grok 4.6 API Access?

LinkModel offers Grok 4.6 at regular token prices 30% below xAI Standard, with no minimum spend or monthly platform fee.

For short-context requests, that means:

- $1.40/M input, versus $2/M at xAI Standard.

- $0.35/M cached input, versus $0.50/M.

- $4.20/M output, versus $6/M.

For long-context requests, LinkModel lists $2.80/M input, $0.70/M cached input, and $8.40/M output, maintaining the same 30% price advantage over the corresponding xAI Standard rates.

What Makes LinkModel Cost-Effective?

The price advantage is particularly meaningful for applications with sustained usage.

At an equivalent xAI Standard token bill of $1,000 per month, LinkModel's published rates reduce the modeled cost to $700.

At $10,000 per month, the modeled saving rises to $3,000 monthly.

These savings apply to the compared model-token charges, not automatically to separately billed tools or services.

Start Using Grok 4.6 for 30% Less

You do not need a large monthly commitment to benefit from lower Grok 4.6 token rates.

With LinkModel, developers can start with pay-as-you-go usage, avoid monthly platform fees, and access Grok 4.6 alongside other supported AI models through a unified API.

Explore LinkModel to access Grok 4.6 at 30% below xAI Standard token prices.

Frequently Asked Questions

Conclusion

Grok 4.6 API pricing starts at $2/M input, $0.50/M cached input, and $6/M output on xAI Standard, but actual costs depend on context length, cache hits, reasoning output, and tool usage. The 200K context threshold is especially important because long-context rates apply across the request rather than only to excess tokens. Developers can lower spending by keeping prompts relevant, reusing cached content, selecting appropriate reasoning effort, and controlling unnecessary tool calls. For a more affordable starting point, LinkModel offers Grok 4.6 token rates 30% below xAI Standard, starting at $1.40/M input and $4.20/M output, with pay-as-you-go billing, no minimum spend, no monthly platform fee, and unified access to multiple AI models.