xAI Grok API token pricing (serverless, per 1M tokens, 2026-10-04 list price)

USD per 1M TOKENS, LIST PRICE, observed 2026-10-04. Every model with a 500k or 1M context window has TWO rate sets, short context and long context (>=200k tokens), and the rule stated on the page is that long-context rates apply to ALL tokens in a request once the prompt crosses the threshold. grok-4.7 (500k): short $2.00 input / $0.50 cached / $6.00 output; long $4.00 / $1.00 / $12.00. grok-4.6 (500k): same as 4.7. grok-4.5 (500k): short $2.00 / $0.30 / $6.00; long $4.00 / $0.60 / $12.00. grok-4.3 and grok-4.20 variants (1M): short $1.25 / $0.20 / $2.50; long $2.50 / $0.40 / $5.00. grok-build-0.1 (256k): short $1.00 / $0.20 / $2.00; long $2.00 / $0.40 / $4.00. price_usd = grok-4.7 short-context input. Requests to the US regional endpoint are billed at 1.1x the global token rates, a stated 10% premium PERISHABLE-FIGURE STAMP (added 2026-10-04, uniform across data/supply/ so it is greppable). PRICE CLASS: PUBLISHED LIST PRICE - an asking rate a new customer is quoted, NOT a realised, discounted or negotiated transaction rate. A real bill is at or below this and typically well below it once commitments, credits and committed-use discounts apply. OBSERVED: 2026-10-04 RE-VERIFY BY: 2027-01-04 - the repo's quarterly re-verification bar. After that date this figure is presumed drifted and must be re-checked against its source before it is quoted again. A price without its observation date is noise; this stamp exists so a reader can tell a fresh figure from a stale one without reading prose. UNIT: USD per 1M TOKENS, LIST PRICE, observed 2026-10-04 READING RULE: No further perishable flag is published for this figure.

lead_time_weeks is null because NO NUMBER APPLIES, and the reason is specific rather than generic: this record prices a metered service term (tokens, egress, storage) rather than hardware capacity, so there is no order-to-delivery interval to report at all. Calling this 'no published lead time' would be wrong in the other direction - nothing was withheld.

no EAR citation: no controlled item is transferred by this transaction; not an EAR citation: provider service terms and end-use policy (the AI Diffusion Rule briefly tiered service access before its 2025-05-13 rescission)

MECHANISM: a token-priced API, so no accelerator crosses a border and the 3A090 hardware licence regime does not attach to the transaction. Export control becomes an END-USER AND TERMS question instead: the provider applies geographic and end-use restrictions to service access, and a customer can still be refused under sanctions or end-use policy wherever the compute sits. A reader targeting a restricted destination should read the provider's terms, not the EAR.

xAI's long-context rule is the harshest of the three providers that have one (OpenAI 2x, Anthropic n/a per-token, xAI 2x) because it is ALL-OR-NOTHING on the whole request, not a per-token split: one token past 200,000 doubles the entire bill including the tokens that were already short-context priced. Any xAI cost model that multiplies tokens by a blended rate will be wrong at the boundary. The US endpoint premium is a real 10% and it is easy to forget because the endpoint choice is a URL, not a plan. Non-text charges on the same page for completeness: image generation from $0.002-$0.01/img at $0.02-$0.25 out depending on resolution and model; video from $0.002/img + $0.01/sec input; speech-to-text $0.10/hr REST and $0.20/hr streaming; TTS $15.00 per 1M characters; speech-to-speech $0.08/min. No GPU hourly rate published. EXPORT CONTROL (2026-10-04): NOT APPLICABLE IN THE HARDWARE-SHIPMENT SENSE - this record prices a token-based API, so no accelerator is transferred across a border and the ECCN 3A090 hardware licence regime does not attach to the transaction. Recorded null rather than false precisely because a token API is where export control instead becomes an END-USER/TERMS question: the provider applies geographic and end-user restrictions to service access (the AI Diffusion Rule briefly put a tiered country structure on exactly this kind of service before being rescinded on 2025-05-13), and a customer can still be refused under sanctions or end-use policy regardless of where the compute sits. Anyone reading this record for a restricted destination should look at the provider's terms, not at the EAR. Sources for the regime: spl-bis-2025-05-ai-diffusion-rescission, spl-bis-2026-05-d5-guidance. EXPORT CONTROL (2026-10-04): NOT APPLICABLE IN THE HARDWARE-SHIPMENT SENSE - this record prices a token-based API, so no accelerator crosses a border and the ECCN 3A090 hardware licence regime does not attach to the transaction. Recorded NULL rather than false precisely because a token API is where export control instead becomes an END-USER and TERMS question: the provider applies geographic and end-user restrictions to service access (the AI Diffusion Rule briefly put a tiered country structure on exactly this kind of service before being rescinded and unenforced from 2025-05-13), and a customer can still be refused under sanctions or end-use policy regardless of where the compute sits. For a restricted destination, read the provider's terms, not the EAR. Sources: spl-bis-2025-05-ai-diffusion-rescission, spl-bis-2026-05-d5-guidance. LEAD TIME: null - correct and not a gap. A token-priced serverless API is provisioned on request, so there is no lead time to publish; the binding constraint when a model is at capacity is provider-side queuing and any per-account concurrency limit, which this repo records in notes where a vendor publishes one (see the DeepSeek concurrency figures already here). Deliberately NOT recorded: any 'model capacity lead time' estimate, because vendors publish none and the SEO 'GPU lead time' numbers that circulate do not apply to a rented endpoint. Probe log in source spl-negative-no-oem-publishes-lead-time. VOLATILE STRUCTURE, RECORDED WHILE IT IS STILL TRUE - observed 2026-10-04. Same reasoning as the DeepSeek record: the durable fact is the SHAPE of the pricing, and the shape is what will change. AS OF 2026-10-04 every xAI model with a 500k or 1M context window publishes TWO rate sets, short-context and long-context, and the threshold is 200,000 tokens. THE RULE AS PUBLISHED IS ALL-OR-NOTHING: once a request's prompt crosses 200k, the long-context rate applies to ALL TOKENS IN THAT REQUEST, including the tokens that would otherwise have been short-context priced. It is not a per-token split and it is not a marginal surcharge on the tail. One token past the threshold doubles the entire bill. THE MEASURED EFFECT, not the rhetorical one: exactly 2x across every column of every affected model - grok-4.7 and 4.6 short $2.00 in / $0.50 cached / $6.00 out becoming $4.00 / $1.00 / $12.00; grok-4.5 $2.00 / $0.30 / $6.00 -> $4.00 / $0.60 / $12.00; grok-4.3 and the 4.20 variants at a 1M window $1.25 / $0.20 / $2.50 -> $2.50 / $0.40 / $5.00; grok-build-0.1 at 256k $1.00 / $0.20 / $2.00 -> $2.00 / $0.40 / $4.00. A per-token blended rate is therefore wrong AT THE BOUNDARY, not approximately wrong. WHY IT WILL STALE: a 2x cliff is a scarcity-pricing instrument, and scarcity pricing is the first thing to be revised when capacity arrives. Both the threshold and the multiplier are uncommitted. Note also that xAI and OpenAI both double at >=200k today - two independent providers converging on the same threshold suggests an industry convention rather than a coincidence, which is itself worth knowing before either changes it. SECOND VOLATILE FACT ON THE SAME PAGE, recorded because it is a price multiplier that is easy to miss: requests to the US regional endpoint are billed at 1.1x the global token rates. The endpoint choice is a URL rather than a plan, so a model can be 10% wrong purely from configuration, and nothing in the token count would reveal it. No GPU-hour rate is published.