← All providers

xAI API pricing

5 xAI models tracked, updated daily. Full rates below, with guidance on which tier fits which workload.

ModelTierInputOutputContext
Grok Build 0.1Standard$1.00$2.00256K
Grok 4.20Standard$1.25$2.502M
Grok 4.20 Multi-AgentStandard$1.25$2.502M
Grok 4.3Standard$1.25$2.501M
Grok 4.5Standard$2.00$6.00500K

Cached input rates and full detail on the pricing page. Prices in USD per 1M tokens.

xAI runs a more concentrated lineup than most providers — the current roster sits entirely in the Standard tier, spanning the numbered Grok releases (4.3, 4.5) alongside the "Build" preview line and a "Multi-Agent" variant. See the live table above for current rates.

No Budget or Premium tier — what that means for routing

Because Grok doesn't currently split into distinct Budget/Standard/Premium bands the way OpenAI or Google do, the usual "route simple tasks to a cheap tier" strategy (see How to Cut Your LLM API Bill by 80%) applies differently here: within the Grok lineup itself there's less headroom to downgrade for simple tasks. If your workload has a meaningful share of low-complexity requests, it's often worth comparing a Grok model against a genuinely Budget-tier model from another provider — DeepSeek, Qwen, or Moonshot — rather than assuming you need to stay within one provider's lineup.

The Multi-Agent variant

The Multi-Agent-branded Grok model is positioned specifically for orchestration and multi-step agentic workflows. If you're building an agent that coordinates multiple sub-tasks or tool calls, it's worth checking this variant against the standard numbered releases directly on the compare page — agentic workloads are also exactly where prompt caching tends to pay off most, since the tool schema and orchestration instructions repeat on every step.

Newer releases and version churn

Grok's numbered releases move relatively quickly, and a "Build"-branded preview line sits alongside the numbered models — preview and build variants can have different pricing and stability characteristics than the numbered releases, so confirm you're looking at the variant you intend to ship with, not a preview snapshot, before committing to it in production.

Coding and reasoning performance

Check Grok's standing on the LiveBench and SWE-bench Verified cost-vs-quality charts before assuming it's the right fit for a coding-heavy workload — benchmark rank shifts between releases, and the value-frontier view will tell you whether a given Grok release is currently a strong price-for-capability pick or dominated by a cheaper alternative.

When xAI vs. a competitor

With the lineup concentrated in a single tier, the comparison against competitors is usually straightforward: check Grok against the Standard-tier models from OpenAI, Google, and Anthropic on the compare page for your specific workload, and don't overlook the lower-cost providers if a meaningful share of your traffic is simple enough to not need Standard-tier capability at all.

See current Grok pricing live above, or put it head-to-head with any competitor on compare.

Frequently asked questions

How many xAI models does TokenCost track?
TokenCost currently tracks 5 xAI models, refreshed daily from live pricing data.
What is the cheapest xAI model?
Grok Build 0.1 is currently xAI’s lowest-cost model on input price, at $1.00 per 1M input tokens and $2.00 per 1M output tokens.
Does xAI support prompt caching?
Where a cached-input rate is published, TokenCost shows it directly in the pricing table (cached input is typically billed at a steep discount versus the standard input rate). See our prompt caching guide for the general mechanics.