xAI Grok API Review 2026

What you’re buying from xAI

The xAI API gives programmatic access to the Grok family of models through console.x.ai, with OpenAI and Anthropic compatible endpoints plus a native SDK. xAI launched the API in October 2024, and its release notes now describe Grok as a SpaceXSI model, reflecting the SpaceX ownership structure announced earlier this year. The product lines are text models (Grok 4.7 down to grok-4.3 and the 4.20 family), the Imagine API for images and video, and voice models for speech in and out.

For a developer the unit that matters is the team. Credits, API keys, rate limit tiers and billing settings all live on a team, and you should confirm which team is active before changing anything under Billing. That design makes it easy to isolate a prototype from production. It also makes it easy to buy credits into the wrong team, which is the support ticket nobody wants to write.

Why there’s no free trial anymore

There’s no free trial, and older guides that say otherwise are out of date. At the December 2024 beta xAI gave every account $25 of monthly credit, and during 2025 its docs described a $150 monthly credit for teams that opted into sharing API data in eligible countries after spending at least $5. Neither appears anywhere in xAI’s docs or console today, and the quickstart simply tells you to load credits before you start.

Credits today are prepaid dollars bought under Billing in console.x.ai. The Buy credits flow runs through Guest Checkout, bank transfers clear in two to three business days, and the same card holds a promo code field for codes xAI issues directly. Auto top-up lets you set a trigger balance, a top-up amount of at least $5 and a monthly ceiling, with a warning at 80% of that ceiling. Set the ceiling. It’s the only brake on the account.

How the rate limit ladder works

The API has no always-free allowance. The only free usage xAI mentions anywhere in its developer docs sits inside Grok Build, its coding product, which has a free tier that excludes the Grok 4.7 Fast model. That usage is separate from the API and can’t be redirected to your own application.

Quotas are rate limits per model on two dimensions, requests per second and tokens per minute. The per-second figure is the per-minute request budget divided by 48 with a floor of two, so a team can’t burn a whole minute of requests in one second. Tier 0, the default, allows 150 requests per second and 50 million tokens per minute on Grok 4.7, 4.6 and 4.5, and 37 requests per second with 10 million tokens per minute on grok-4.3, the 4.20 models and grok-build. Reasoning tokens and cached tokens both count toward the token limit.

Tiers are earned by cumulative spend since January 1, 2026: Tier 1 at $50, Tier 2 at $250, Tier 3 at $1,000 and Tier 4 at $5,000, where Grok 4.7 reaches 500 requests per second and 100 million tokens per minute. Once you reach a tier you keep it. Exceeding any limit returns a 429, and xAI suggests exponential backoff rather than a quota request as the first fix. Spending $50 early just to clear Tier 0 is a reasonable move for a team with a launch date.

What each Grok model costs you

Grok 4.7 and Grok 4.6 cost $2.00 input, $0.50 cached input and $6.00 output per million tokens with a 500k context window. Grok 4.5 has the same rates with $0.30 cached input, and grok-4.3 plus the three grok-4.20 variants cost $1.25, $0.20 and $2.50 with a 1M window. grok-build-0.1 is $1.00 input and $2.00 output at 256k. Every text model doubles its rates for the whole request once the prompt reaches 200k tokens.

Beyond tokens, tool calls bill at $5 per 1,000 for web search, code execution and file attachments, $2.50 per 1,000 for collections search, and X search charges per post or profile returned. Priority processing doubles the text rate when the response confirms the priority tier. Image generation runs $0.02 to $0.05 per image and video $0.02 to $0.08 per second. Voice is $0.08 per minute for speech to speech and $15 per million characters for text to speech.

Billing is prepaid by default. Monthly invoicing exists but ships with a $0 limit, which means requests stop the moment prepaid credits run out until you ask sales and raise the limit. Invoices appear under Billing and can’t be regenerated, and xAI won’t let you delete the last payment method on file. The Usage explorer and the cost_in_usd_ticks field on every response are your two tools for tracking spend, and the second one is the more useful.

Why founders shouldn’t wait for a program

xAI runs none. The docs have no page for startups, research credits, education or a cloud marketplace program, and the 2026 news page is about model releases, funding and the Grok Business and Grok Enterprise seat products rather than API grants. Several accelerator perk portals list Grok credits, but those are partner arrangements with no matching xAI page, so we don’t list them as offers.

What xAI does offer organizations is an enterprise route. Capacity beyond Tier 4, Imagine API limit increases and monthly invoiced billing all go through [email protected], and you can also request a higher limit inside the console without extra spend. That’s a negotiation, not a program, and it comes with no published credit amount. If you need free usage to validate an idea, do it on another provider and bring the winning prompts here.

Which models actually change the bill

Grok 4.7, released September 21, 2026, is the model to benchmark first for coding and agent work, and it ships at the same $2 and $6 rates as its predecessor. For high-volume text, grok-4.3 at $1.25 and $2.50 with 1M context is the workhorse, and grok-4.20-multi-agent-0309 at the same price runs parallel sub-agents inside one request. The Batch API covers all of these at 20% off and now accepts image and video generation and JSONL uploads of up to 50,000 requests.

Two features change the cost picture. Prompt caching has been available since May 2025 and cuts input cost by 75% or more on stable prefixes, which for a chat app with a long system prompt is most of the input bill. The US regional endpoint, available for Grok 4.7 and 4.6, keeps processing in the United States at 1.1 times the global rate, which matters only if a contract says it does.

Where the money leaks

Support for self-serve teams is the docs, the console and an email to sales for billing changes, with no published support SLA below enterprise. The risks we’d flag are financial rather than technical. Auto top-up without a monthly maximum can keep buying credits during a runaway loop, and the 80% warning only fires against a limit you set. Flagged requests are still billed after generation, and a request blocked before generation in the Responses API costs $0.05.

Account verification is lighter than on cloud providers because there’s nothing free to abuse, but Guest Checkout means card details are entered at purchase rather than stored up front. Rate tier progress is tied to the team, so deleting a team and starting over resets you to Tier 0, and the $5,000 you spent to reach Tier 4 counts for nothing.

Who should pay for Grok

A funded team that has already picked Grok on quality and wants the lowest frontier output price with a 20% batch discount on top. Not anyone who needs free usage to validate an idea, because every token costs money from the first call and no program changes that. 3.4 out of 5. Strong pricing and a clean rate tier ladder, held back by the complete lack of credits and a $0 invoiced default that surprises teams the first time the balance hits zero.