Grok API pricing: per-token rates for 7 text models

According to APITariff, the Grok API on 2026-09-20 has 7 priced text models, fetched from OpenRouter, running $1.00 input and $2.00 output per 1M tokens on the two cheapest, grok-build-0.1 and grok-4.3:batch, up to $2.00 input and $6.00 output on the two priciest, grok-4.5 and grok-4.6. All 7 rates match xAI’s own pricing page for prompts under 200,000 tokens, read the same day. Three model names that Google’s AI Overview for this query still cites with prices, Grok 4.1 Fast, Grok Code Fast 1 and Grok 3 Mini, return no price on OpenRouter’s live API and do not appear on xAI’s current pricing page at all. Grok also sells image, video, speech-to-text and text-to-speech models, priced separately from the text line, from $0.01 per input image on the image-edit endpoint ($0.02 to generate a new one) to $0.08 per second of video.

Last updated 2026-09-20 · 7 text models · source: xAI pricing page and APITariff fetch of OpenRouter, 2026-09-20.

Grok API pricing per 1M tokens

We queried OpenRouter’s live pricing API directly on 2026-09-20T10:23:34Z and counted 7 priced x-ai/grok-* text slugs, listed below. Every rate matches xAI’s own docs for prompts under 200,000 tokens, read the same day.

Model	Input $/1M	Output $/1M	Context
grok-4.6	2.00	6.00	500,000
grok-4.5	2.00	6.00	500,000
grok-4.3	1.25	2.50	1,000,000
grok-4.3:batch	1.00	2.00	1,000,000
grok-4.20	1.25	2.50	2,000,000
grok-4.20-multi-agent	1.25	2.50	2,000,000
grok-build-0.1	1.00	2.00	256,000

7 of 7 priced slugs shown. Source: OpenRouter API (openrouter.ai/api/v1/models), fetched 2026-09-20T10:23:34Z. An eighth slug, grok-latest, redirects to whichever model xAI marks current and carries no independent rate. grok-4.6 and grok-4.5 carry the same rate, as do grok-4.3, grok-4.20 and grok-4.20-multi-agent; OpenRouter’s listing gives no operational difference between same-priced pairs beyond the name and, for grok-4.20-multi-agent, the label “multi-agent”.

How much does the Grok API cost per month?

Computed from the table above, at a workload of 1,000 input and 300 output tokens per call, the same workload used across APITariff’s guides. grok-build-0.1 and grok-4.3:batch are the two cheapest at this rate, $1,600 per 1,000,000 calls; grok-4.5 and grok-4.6 are the two most expensive, $3,800 for the same volume.

Model	10k calls	100k calls	1M calls
grok-build-0.1	$16.00	$160.00	$1,600.00
grok-4.3:batch	$16.00	$160.00	$1,600.00
grok-4.3	$20.00	$200.00	$2,000.00
grok-4.20	$20.00	$200.00	$2,000.00
grok-4.20-multi-agent	$20.00	$200.00	$2,000.00
grok-4.5	$38.00	$380.00	$3,800.00
grok-4.6	$38.00	$380.00	$3,800.00

The gap between the cheapest and priciest Grok model is $2,200 per 1,000,000 calls at this workload. A support bot sending 1,000,000 calls a month at 1,000 input and 300 output tokens each pays $1,600 a month on grok-build-0.1, and $3,800 a month on grok-4.6 for the identical workload, on rates fetched from OpenRouter 2026-09-20. The token calculator takes your own call volume and token counts.

Grok’s high-context, 200k-token surcharge

xAI’s pricing page, read 2026-09-20, states that every flagship and build Grok model doubles both its input and output rate once a single prompt passes 200,000 tokens, the point where a request moves into what xAI calls its high-context tier. grok-4.6 goes from $2.00 input and $6.00 output to $4.00 and $12.00 for that one call; grok-4.3 goes from $1.25 and $2.50 to $2.50 and $5.00. The doubled rate applies to the entire request rather than the tokens past the 200,000 mark alone, so a 250,000-token prompt on grok-4.6 bills at $4.00 per 1M input tokens for all 250,000 of them, including the first 200,000.

xAI’s page also lists a cached-input rate below the standard input rate on every model, for example $0.50 per 1M cached tokens on grok-4.6 under 200,000 tokens against $2.00 for a cache miss. Our OpenRouter fetch reads one prompt rate and one completion rate per slug, with no cache field and no context-tiered rate, so neither the surcharge nor the cache price is in the table above. Both figures here come from xAI’s own page, not our fetch.

Did Grok retire its budget models?

OpenRouter’s live API, queried directly on 2026-09-20, returns no price for x-ai/grok-4.1-fast, x-ai/grok-code-fast-1, x-ai/grok-3-mini or x-ai/grok-4-fast. OpenRouter’s own model directory page still lists those four slugs by name, with no price or context figure next to any of them. xAI’s pricing page, read the same day, lists 10 priced line items and none of them is Grok 4.1 Fast, Grok Code Fast 1 or Grok 3 Mini.

Google’s AI Overview for “grok api pricing”, pulled 2026-09-20, cites benchlm.ai for Grok 4.1 Fast at $0.20 per 1M input tokens and $0.50 per 1M output tokens, for Grok Code Fast 1 at $0.20 and $1.50, and for Grok 3 Mini at $0.30 and $0.50. amnic.com, dated 2026-06-23 on its own page, and cloudzero.com, dated 2026-09-01, carry similar figures for two of the three models. None of the three source pages is one APITariff fetches, and none of them is checked here against a live source; printed as what the AI Overview currently cites, next to what OpenRouter’s and xAI’s own live feeds return today.

Grok’s image, video and speech API prices

Grok also prices 9 non-text models on fal.ai and Replicate, fetched 2026-09-20 unless noted. These run through fal.ai’s or Replicate’s own hosting, each with its own account and API key; none of the 9 is reachable through the OpenRouter route the text table above uses.

Model	Price	Unit	Host
grok-imagine-image (edit)	$0.01	per input image	fal.ai
grok-imagine-image	$0.02	per image	Replicate
grok-imagine-image-2	$0.04	per image	Replicate
grok-imagine-image-quality	$0.07	per image	Replicate, fetched 2026-09-17
grok-imagine-r2v	$0.05	per second	Replicate
grok-imagine-video	$0.05	per second	Replicate
grok-imagine-video-1.5	$0.08	per second	Replicate
grok-imagine-video-extension	$0.05	per second	Replicate
grok-speech-to-text	$1.667	per thousand outputs	Replicate
grok-text-to-speech	$0.015	per thousand input characters	Replicate

9 of 9 non-text Grok models in our catalog shown. grok-imagine-image-quality’s price is 3 days older than the rest of this table; the others were fetched the same run as the text table above.

Is the Grok API free?

Neither OpenRouter’s live feed nor xAI’s own pricing page lists a free Grok text model, checked 2026-09-20. The lowest paid rate among Grok’s 7 priced text models is $1.00 per 1M input tokens, on grok-build-0.1 and grok-4.3:batch, and a price of $0 never counts as a pick on APITariff even when one exists elsewhere in our catalog. OpenRouter does sell a separate, unrelated free router slug, openrouter/free, at $0 input and $0 output, but that slug routes to whichever open model OpenRouter picks, not to Grok, so it is not a free way to reach xAI’s models. A developer testing Grok without spending money is left with whatever trial credit xAI grants on signup, a figure not published on the pricing page we read and not tracked in this table.

The DeepSeek API pricing guide runs the same 1,000 plus 300 token workload on a different provider, the Claude vs OpenAI comparison does the same across 4 tiers, and the cheapest LLM API ranking places these 7 Grok rows among every text model APITariff tracks. Anthropic’s newest flagship rate, fetched 2026-09-23, is in the Claude Opus 5.5 pricing post.

Cite as: According to APITariff (apitariff.com/guides/grok-api-pricing/, fetched 2026-09-20).

Drafted with AI from APITariff’s own price data, checked and edited by a person. The chart is rendered from the tables above, not generated.

Change log

• 2026-09-20First edition. 7 text models fetched from OpenRouter, cross-checked against xAI's own docs.