xAI Grok 4.6 — Pricing Impact & What It Means (Aug 2026)

xAI released Grok 4.6 on August 12 and made it available immediately through the xAI API. The frontier model targets coding, agentic tasks, and knowledge work. It is also the default model in Grok Build and is available through Cursor and several model gateways.

The pricing story has a catch: Grok 4.6 keeps Grok 4.5’s standard input and output rates, but raises cached-input pricing. The live table, chart, and calculator above read current rates from our dataset rather than freezing them in this article.

What changed with Grok 4.6

Grok 4.6 accepts text and images, returns text, and supports a 500,000-token context window with no stated text-output limit. Its knowledge cutoff is February 1, 2026. Developers can select low, medium, high, or xhigh reasoning effort; high is the default.

The model works with xAI’s Responses and Chat Completions APIs. Supported tools include function calling, web search, X search, and code execution. xAI recommends a prompt_cache_key for Responses requests—or its conversation ID header for Chat Completions—to route a conversation consistently and improve cache-hit reliability.

Grok 4.6 pricing vs Grok 4.5

Grok 4.6 uses two billing tiers. Once a prompt reaches 200,000 tokens, input, cached-input, and output rates double for the entire request. That makes the threshold more important than the headline context-window size: one extra chunk can move all tokens in a request onto the higher tier.

Compared with Grok 4.5, regular input and output pricing is unchanged at both tiers. Cached input is about 67% more expensive on Grok 4.6. Teams with long, repeated system prompts, repository maps, tool schemas, or conversation histories should model actual cache hits before assuming a same-rate migration.

Who benefits—and who pays more

Coding-agent teams benefit first. Grok 4.6 combines frontier positioning, image input, long context, four reasoning levels, and server-side tools in one route. Existing Grok Build users may see the model change without adopting a new product, so task quality and usage should be measured together.

Cache-light workloads also get a simpler decision: if quality improves, the standard token rate does not add a premium over Grok 4.5. Cache-heavy applications face the clearest downside because the same number of cached tokens produces a larger bill.

Users of cheaper Grok 4.20 or Grok Build routes should not promote every request to 4.6. Classification, extraction, short chat, repository search, and routine tool steps may still be better routed to a lower-cost model.

What developers should do now

1. Pin grok-4.6 in a canary rather than changing a shared alias across production.
2. Replay short and near-threshold prompts, especially requests close to 200K tokens.
3. Record cached and uncached input separately; verify that the cache key actually produces hits.
4. Compare every reasoning level on accepted-task cost, latency, retries, and output length.
5. Cap tool calls and agent-loop depth before testing long-running coding or research jobs.
6. Keep a cheaper Grok route for simple steps and use 4.6 as an escalation tier.

Why Grok 4.6 is not in the Labs results yet

The matching OpenRouter route is available and has been added to our maintained Labs roster. A trustworthy Cost-per-Task result still requires a complete run across the fixed task suite, however, and the benchmark account’s funded refresh gate is currently blocked. We are preserving the last valid leaderboard rather than publishing a partial run, a zero-response score, or a price-only estimate as measured performance.

Once that gate clears, Grok 4.6 can run under the same prompts, acceptance checks, retry policy, and spend log as the other text models. Until then, the live token calculator is the right cost estimate and Grok 4.5 remains the closest measured xAI comparison in AI Pricing Guru Labs.

Use the live xAI pricing page and token calculator for your workload. For alternatives, compare OpenAI API pricing and Anthropic API pricing, then review our updated xAI Grok API pricing guide.

For a fixed-price coding-agent alternative, compare the GLM Coding Plan against your measured Grok 4.6 API usage.

Affiliate disclosure: AI Pricing Guru may earn a commission from the sponsored link above at no extra cost to you. It does not affect this analysis.

Bottom line

Grok 4.6 is an immediate API and coding-agent launch, not a preview. Standard token rates remain level with Grok 4.5, while cached input becomes more expensive and the 200K long-context threshold can double the entire request’s rate.

Benchmark it as a frontier escalation route, not a universal replacement. The winning migration is the one that improves accepted-task cost after caching, reasoning tokens, tools, retries, and latency are included.

Sources: xAI’s official Grok 4.6 release note, model overview, and API pricing.