xAI API Pricing, Models & Rate Limits (2026)

1. Regional and long-context tariff ledger

Region is an eligibility field, not a quality signal. A long-context amount is shown only as the fixed token formula; threshold and regional parity remain separate.

2. Tool-enabled invoice surface

Tool units are not folded into token cost. Missing xAI unit prices remain Unavailable.

3. Versioned-model rollout canary

Reasoning control and structured output/tool support are version-specific evidence fields; availability never substitutes for quality evidence.

Verified 2026-08-14. Data owner: Luna. “Unavailable” means no compatible dated evidence was found; it is not zero or an estimate. Re-verify dated rates, specs, and policy before production use. First-party source · Run this scenario →

1. Regional failover budget

Formula: failover spend = primary bill + (shadow share × duplicate compatible bill). Outage likelihood and SLA are not derived from region eligibility or price.

2. Live-tool inverse budget

3. Portfolio output-cap guardrail

Guardrail formula: output spend = input rate × fixed input + output rate × capped output. Reasoning-control support and quality remain independent evidence fields.

Verified 2026-08-14. Data owner: Luna. “Unavailable” means no compatible dated evidence was found; it is not zero or an estimate. Source / registry · Run this scenario →

1. OpenAI-SDK conformance-drift canary

Formula / rule: drift=field/path/value difference under identical request Source: pricing registry verified 2026-08-26. Rate: Grok 4.6, $2.0000 input/M + $6.0000 output/M.

2. Streamed cancellation and usage-reconciliation ledger

Formula / rule: cost=returned usage+restart; missing invoice is not zero Source: pricing registry verified 2026-08-26. Rate: Grok 4.6, $2.0000 input/M + $6.0000 output/M.

3. Multi-hop source-disagreement audit

Formula / rule: accepted=claim support∧resolution or uncertainty Source: pricing registry verified 2026-08-26. Rate: Grok 4.6, $2.0000 input/M + $6.0000 output/M.

1. Reasoning-effort-to-latency/cost tradeoff ledger

Formula / rule: Final-answer-only floor cost = (frozen prompt tokens × input rate + final-answer tokens × output rate)/1M at the Grok 4.6 registry rate, excluding the unobserved reasoning-token count. Actual reasoning-token consumption, wall-clock latency, and final-answer changes across effort levels require a matched run, which is not present in the registry, so total cost including reasoning tokens is Unavailable at every level. Source: pricing registry verified 2026-08-26.

2. Function-calling schema-rejection-and-repair-loop audit

Formula / rule: Retry-budget cost = frozen-fixture token bill at the Grok 4.6 registry rate. Initial rejection reason, self-correction attempts, and resolved-versus-abandoned outcome require a matched run, which is not present in the registry, so only the retry-budget cost below is reproducible. Source: pricing registry verified 2026-08-26.

3. X-platform live-search citation freshness-decay ledger

Formula / rule: Per-checkpoint query cost = fixed-query token bill at the Grok 4.6 registry rate, re-run at 0/24/72 hours after the initial answer. Cited-post age drift, superseded or retracted claims, and citation churn require a matched re-run and a sourced staleness-detection method, neither of which is present in the registry, so only the per-checkpoint query cost below is reproducible. Source: pricing registry verified 2026-08-26.

1. Structured-output (JSON mode) schema-compliance audit

Formula / rule: Base-request cost = (frozen prompt+schema tokens × input rate + response tokens × output rate)/1M at the Grok 4.6 registry rate. Declared-mode acceptance, malformed-output rate, repair-call count, and token overhead versus an equivalent free-form completion require a matched run, which is not present in the registry, so only the base-request cost below is reproducible. Source: pricing registry verified 2026-08-26.

2. Multi-key/team spend-attribution ledger

Formula / rule: Base-sequence cost = frozen fixed-key-set token bill at the Grok 4.6 registry rate for one billed account. Whether the documented usage endpoint attributes cost per key, per project, or only in aggregate requires a sourced usage-endpoint schema document, which is not present in the registry, so attribution granularity below is Unavailable and named rather than assumed. Source: pricing registry verified 2026-08-26.

3. Context-window overflow truncation-policy audit

Formula / rule: Base-request cost = frozen near-limit/over-limit prompt token bill at the Grok 4.6 registry rate, priced up to the documented context-window ceiling. The documented truncation rule (reject versus silently truncate, and which end), observed behavior, and cost impact of an unexpected silent truncation require a matched near/over-limit run, which is not present in the registry, so only the priced-up-to-ceiling cost below is reproducible. Source: pricing registry verified 2026-08-26.

1. Live Search per-source-type cost-attribution ledger

Formula / rule: Base-request cost = (frozen prompt tokens × input rate + response tokens × output rate)/1M at the Grok 4.6 registry rate, excluding any Live Search per-source charge. A documented per-source-type (web, news, X-post) charge breakdown for a Live Search call requires a matched Live Search invoice, which is not present in the registry, so only the base-request cost below is reproducible. Source: pricing registry verified 2026-08-26.

2. System-versus-user token-accounting parity ledger

Formula / rule: Combined-prompt cost = (frozen system-prompt tokens + user-prompt tokens) × input rate/1M plus response tokens × output rate/1M at the Grok 4.6 registry rate, treating both roles as one billed input pool. Whether the registry bills system-role and user-role prompt tokens at the same per-token rate or applies a role-specific differential requires a sourced role-differentiated rate card, which is not present in the registry, so parity below is Unavailable and named rather than assumed. Source: pricing registry verified 2026-08-26.

3. Spend-milestone rate-limit-escalation ledger

Formula / rule: Base-request cost = frozen-request-sequence token bill at the Grok 4.6 registry rate for the stated volume. Whether crossing a documented cumulative-spend milestone changes requests-per-minute or tokens-per-minute ceilings requires a sourced escalation-threshold document, which is not present in the registry, so only the base-request cost below is reproducible. Source: pricing registry verified 2026-08-26.

1. Live Search per-source-type cost-attribution ledger

Formula / rule: Base-request cost = (frozen prompt tokens × input rate + response tokens × output rate)/1M at the Grok 4.6 registry rate, excluding any Live Search per-source charge. A documented per-source-type (web, news, X-post) charge breakdown for a Live Search call requires a matched Live Search invoice, which is not present in the registry, so only the base-request cost below is reproducible. Source: pricing registry verified 2026-08-26.

2. System-versus-user token-accounting parity ledger

Formula / rule: Combined-prompt cost = (frozen system-prompt tokens + user-prompt tokens) × input rate/1M plus response tokens × output rate/1M at the Grok 4.6 registry rate, treating both roles as one billed input pool. Whether the registry bills system-role and user-role prompt tokens at the same per-token rate or applies a role-specific differential requires a sourced role-differentiated rate card, which is not present in the registry, so parity below is Unavailable and named rather than assumed. Source: pricing registry verified 2026-08-26.

3. Spend-milestone rate-limit-escalation ledger

Formula / rule: Base-request cost = frozen-request-sequence token bill at the Grok 4.6 registry rate for the stated volume. Whether crossing a documented cumulative-spend milestone changes requests-per-minute or tokens-per-minute ceilings requires a sourced escalation-threshold document, which is not present in the registry, so only the base-request cost below is reproducible. Source: pricing registry verified 2026-08-26.

1. Cached-input eligibility and hit/miss invoice ledger

Formula / scoring rule: Base bill = registry token bill for the fixed prefix/repeat fixture. xAI cache rate, key scope, retention, and mutation invalidation require a dated matched invoice; base text rates never substitute. Source: pricing registry verified 2026-08-27.

2. Image-generation endpoint size/count/format billing reconciliation

Formula / scoring rule: Text baseline is registry-computed; total accepted-image cost additionally requires per-image/token units, rejection/retry records, delivered dimensions, and a dated generation rate. Source: pricing registry verified 2026-08-27.

3. Function-result payload token-attribution audit

Formula / scoring rule: Base bill = combined declared input plus output tokens at the Grok 4.6 registry rate. Schema, assistant-call, tool-result, reasoning, truncation, repair, and acceptance fields must be joined from one run. Source: pricing registry verified 2026-08-27.

1. Completed streaming-versus-non-streaming usage-parity ledger

Formula / scoring rule: Parity = accumulated chunks + terminal usage compared with non-stream usage for identical completed requests. Exact bill uses returned reasoning/input/output fields; client cancellation is outside this module. Source: pricing registry and dated evidence index verified 2026-08-27.

2. Rate-limit header calibration audit under fixed worker ramps

Formula / scoring rule: Calibration error = advertised remaining/reset or `Retry-After` − observed acceptance recovery. Spend is computed from completed returned usage; header values do not substitute for an observed ramp. Source: pricing registry and dated evidence index verified 2026-08-27.

3. Timeout retry and request-ID duplicate-charge canary

Formula / scoring rule: Resolved-result cost = invoice lines attributable to one server completion ÷ accepted resolved result. A retry is not free and duplicate suppression is not assumed without request-ID/idempotency evidence. Source: pricing registry and dated evidence index verified 2026-08-27.

1. n=1/2/4 multi-choice acceptance and bill ledger

Formula / scoring rule: Total = shared prompt usage + Σ(choice reasoning/output usage) + repair usage; qualify only accepted choices with supported parameter and returned usage fields. Source: pricing registry and dated evidence index verified 2026-08-27.

2. Stop-sequence early-termination canary

Formula / scoring rule: Bill returned reasoning/input/output fields through the matched stop; continuation repair is an additional returned output, not a free completion. Source: pricing registry and dated evidence index verified 2026-08-27.

3. Streamed function-call fragment assembly audit

Formula / scoring rule: Accept when chunk index, call ID/name, UTF-8 JSON reconstruction, terminal usage, and duplicate/omitted-call review all pass; cost is bill ÷ accepted tool result. Source: pricing registry and dated evidence index verified 2026-08-27.

1. JSON-Schema keyword-support and repair-cost matrix

Formula / scoring rule: Accepted cost = first-pass usage + repair usage for schemas that pass keyword support and reviewer acceptance; ignored keywords are failures, not passes. Source: pricing registry and dated evidence index verified 2026-08-27.

2. Image URL-versus-data-URI-versus-file/asset-reference accounting canary

Formula / scoring rule: Image bill = returned image/input/reasoning/output usage plus sourced asset units; identical decoded pixels do not imply identical transport cost. Source: pricing registry and dated evidence index verified 2026-08-27.

3. Organization usage-export pagination and UTC-boundary invoice ledger

Formula / scoring rule: Close = joined request IDs across UTC buckets and cursors; residual = export aggregate − invoice total after late adjustments and duplicate suppression. Source: pricing registry and dated evidence index verified 2026-08-27.

1. Web/X domain, handle, date-range, and recency conformance ledger

Formula / scoring rule: Gate = parameter acceptance + source/date fit + citation validity + joined search/model usage; empty results remain visible. Source: pricing registry and dated evidence index verified 2026-08-27.

2. Collection chunking and retrieval-parameter canary

Formula / scoring rule: Cost per accepted answer = ingestion/storage + retrieval/tool + model tokens + retries; grounding score must be joined to chunk/top-k controls. Source: pricing registry and dated evidence index verified 2026-08-27.

3. Parallel-versus-serial function-dispatch ledger

Formula / scoring rule: Accepted cost = definitions + call arguments + result association/resend + reasoning/final usage; dependent calls remain ordered. Source: pricing registry and dated evidence index verified 2026-08-27.

1. Collection replace/delete propagation and residual-charge ledger

Formula / scoring rule: Close = version propagation + stale-chunk check + storage/retrieval/model units + reindex retry + accepted answer + final bill. Source: pricing registry and dated evidence index verified 2026-08-27.

2. Web/X citation canonicalization canary

Formula / scoring rule: Canonicalization record joins requested/final URL or post ID, redirect chain, author/time, claim span, duplicate collapse, usage, reviewer acceptance, and cost; canonical does not mean true. Source: pricing registry and dated evidence index verified 2026-08-27.

3. Previous-response and conversation-state chain integrity ledger

Formula / scoring rule: Accepted cost = resent input + retained/cache/reasoning/output usage + recovery replay; equivalence requires predecessor identity and context integrity. Source: pricing registry and dated evidence index verified 2026-08-27.

1. Live Search result-budget saturation ledger

Formula / scoring rule: Marginal cost = additional search/model units ÷ supported-claim gain; caps, source types, unique sources, duplicate collapse, context, latency, and reviewer acceptance must join. Source: pricing registry and dated evidence index verified 2026-08-27; provider registry: xAI Grok-3 / Live Search registry rate verified 2026-08-27; test suite: Batch 30 xAI search-budget fixture/test suite (run and result recorded 2026-08-27).

2. X reply, quote, repost, and thread reconstruction canary

Formula / scoring rule: Reconstruction = post/author/event IDs + timestamps + traversal depth + missing-node disclosure + claim-span support + repair queries + reviewer acceptance; canonical citation is not thread structure. Source: pricing registry and dated evidence index verified 2026-08-27; provider registry: xAI Grok-3 / posts-and-responses registry rate verified 2026-08-27; test suite: Batch 30 xAI thread-reconstruction fixture/test suite (run and result recorded 2026-08-27).

3. Collection cross-key and cross-project access-isolation ledger

Formula / scoring rule: Isolation = grant/revocation state + retrieval leakage check + returned spans + units + propagation + retry + accepted answer + attributable charge; unrelated credentials must not be presumed isolated. Source: pricing registry and dated evidence index verified 2026-08-27; provider registry: xAI Grok-3 / Collections registry rate verified 2026-08-27; test suite: Batch 30 xAI access-isolation fixture/test suite (run and result recorded 2026-08-27).

1. X handle rename, recycled-handle, suspended, and deleted-account identity ledger

Formula / scoring rule: Identity continuity = stable author/post/event IDs + timestamp order + missing-state disclosure + supported claims; a handle is not a person identifier. First-party registry: allaiask.com pricing and evidence registry, verified 2026-08-27. Provider/model source: xAI Grok-3 X identity and usage registry, verified 2026-08-27.

2. Collection upload-to-index readiness and deletion-tombstone ledger

Formula / scoring rule: Close = upload acknowledgement + chunk/index state + first searchable time + stale-hit canary + replacement/delete propagation + sourced units; readiness is not inferred from upload. First-party registry: allaiask.com pricing and evidence registry, verified 2026-08-27. Provider/model source: xAI Grok-3 Collections and pricing registry, verified 2026-08-27.

3. Live Search pagination/cursor and partial-page settlement audit

Formula / scoring rule: Marginal cost = additional search/model units ÷ newly supported claims; cursor, duplicate IDs, partial errors, latency, and reviewer acceptance must join. First-party registry: allaiask.com pricing and evidence registry, verified 2026-08-27. Provider/model source: xAI Grok-3 Live Search pricing registry, verified 2026-08-27.

1. Live Search automatic-versus-forced-versus-disabled trigger ledger

Formula / scoring rule: Marginal search value = newly supported claims ÷ additional search/model units; missed or unnecessary search is recorded. xAI Grok-3 Live Search trigger and pricing registry. Dated registry and evidence index, verified 2026-08-27.

2. Encrypted-reasoning or opaque-state carry-forward canary

Formula / scoring rule: Continuity = accepted state field + visible/hidden boundary + tool continuity + answer equivalence + returned usage; encryption is not inferred. xAI Grok-3 state and usage evidence registry. Dated registry and evidence index, verified 2026-08-27.

3. Generated-image revision and variation settlement ledger

Formula / scoring rule: Revision cost = submitted/reference units + job attempts + regenerated regions + storage/download units ÷ accepted revisions. xAI Grok-3 image revision and pricing registry. Dated registry and evidence index, verified 2026-08-27.

1. X-post attached-image/video grounding ledger

Formula / scoring rule: Grounding acceptance = stable post/author/media IDs + fetch state + claim-span support + reviewer acceptance; missing media is disclosed. First-party pricing/evidence registry: xAI X search, attached media, and usage evidence, verified 2026-08-27; unsupported units or credits remain Unavailable.. Dated registry and evidence index, verified 2026-08-27. First-party sources: xAI image and video understanding documentationxAI models and pricing.

2. Generated-image prompt-expansion and revised-prompt attribution canary

Formula / scoring rule: Accepted image cost = generation/retry units for the submitted prompt; undisclosed expansion remains Unavailable and is not inferred from pixels. First-party pricing/evidence registry: xAI image generation, moderation, and pricing evidence, verified 2026-08-27; unsupported units or credits remain Unavailable.. Dated registry and evidence index, verified 2026-08-27. First-party sources: xAI image generation documentationxAI models and pricing.

3. Generated-video continuation and keyframe-conditioning ledger

Formula / scoring rule: Cost per accepted second = returned generation/retry/storage/download units ÷ accepted output seconds; conditioning and overlap are explicit. First-party pricing/evidence registry: xAI video generation, keyframes, and pricing evidence, verified 2026-08-27; unsupported units or credits remain Unavailable.. Dated registry and evidence index, verified 2026-08-27. First-party sources: xAI video generation documentationxAI models and pricing.

1. Responses API versus Chat Completions state-and-invoice parity ledger

Formula / scoring rule: Parity = jointly accepted fields + effective model/state + usage + semantic acceptance + invoice match; SDK conformance is not parity evidence. xAI endpoint parity and pricing evidence; dated registry verified 2026-08-27; unsupported units fail closed as Unavailable.. Dated registry and evidence index, verified 2026-08-27. First-party sources: xAI API documentationxAI models and pricing.

2. X Search `from_date`/`to_date` inclusive-boundary and timestamp canary

Formula / scoring rule: Boundary acceptance = requested/resolved UTC range + post/event identity + creation/edit semantics + supported claim + reviewer result. xAI X Search date-boundary evidence; dated registry verified 2026-08-27; unsupported units fail closed as Unavailable.. Dated registry and evidence index, verified 2026-08-27. First-party sources: xAI search documentationxAI models and pricing.

3. Collections rank/score-threshold and metadata-filter marginal-yield ledger

Formula / scoring rule: Marginal yield = relevant returned chunks ÷ returned chunks after rank, threshold, cap, stale, and access gates; collection/model units use returned billing data. xAI Collections retrieval-control evidence; dated registry verified 2026-08-27; unsupported units fail closed as Unavailable.. Dated registry and evidence index, verified 2026-08-27. First-party sources: xAI collections guidexAI models and pricing.

1. Batch support, mixed-row atomicity, expiry, cancellation, and discount ledger

Formula / scoring rule: Batch acceptance = endpoint/model + row terminal states + result association + retry subset + artifact hash + online-equivalent acceptance + bill. xAI Batch API matched settlement ledger; dated first-party registry verified 2026-08-27; unsupported units fail closed as Unavailable. Dated first-party pricing/evidence registry, verified 2026-08-27. Module-local first-party sources: xAI Batch API documentationxAI models and pricing.

2. Multi-turn Live Search result-reuse and duplicate-debit canary

Formula / scoring rule: Reuse acceptance = response/predecessor/search IDs + reused/new source IDs + citation spans + state/cache fields + marginal search bill. xAI Live Search stateful reuse matched canary; dated first-party registry verified 2026-08-27; unsupported units fail closed as Unavailable. Dated first-party pricing/evidence registry, verified 2026-08-27. Module-local first-party sources: xAI search documentationxAI models and pricing.

3. Direct image-input preprocessing and near-duplicate accounting ledger

Formula / scoring rule: Image parity = submitted/decoded dimensions + pixel hashes + preprocessing state + image/input/output usage + accepted result + bill. xAI direct image preprocessing matched ledger; dated first-party registry verified 2026-08-27; unsupported units fail closed as Unavailable. Dated first-party pricing/evidence registry, verified 2026-08-27. Module-local first-party sources: xAI image input documentationxAI models and pricing.

1. Deferred/background response submit, poll, cancel, and expiry ledger

Formula / scoring rule: Deferred acceptance = endpoint/model + request/response IDs + queued/running/terminal state + polling effect + partial side effects + usage + retry/retention + invoice. xAI deferred response matched ledger; dated first-party evidence/pricing registry verified 2026-08-27; unsupported fields fail closed as Unavailable. Dated first-party pricing/evidence registry, verified 2026-08-27. Module-local first-party sources: xAI Responses API documentationxAI model pricing.

2. Collection duplicate-content and reindex identity canary

Formula / scoring rule: Index identity = file/content hash + collection/chunk/index version + dedup/replacement + stale visibility + ingestion/storage/retrieval units + scope + lag + bill. xAI collection reindex matched canary; dated first-party evidence/pricing registry verified 2026-08-27; unsupported fields fail closed as Unavailable. Dated first-party pricing/evidence registry, verified 2026-08-27. Module-local first-party sources: xAI collections documentationxAI model pricing.

3. Realtime voice session-update boundary and settlement ledger

Formula / scoring rule: Voice mutation = session/event/response IDs + accepted/effective configuration + buffered audio/context + duplicate effects + modality usage + recovery + latency + charge. xAI Realtime voice mutation matched ledger; dated first-party evidence/pricing registry verified 2026-08-27; unsupported fields fail closed as Unavailable. Dated first-party pricing/evidence registry, verified 2026-08-27. Module-local first-party sources: xAI Realtime documentationxAI model pricing.

1. Simultaneous Web Search and X Search arbitration ledger

Formula / scoring rule: Search arbitration = effective modes + call order + source/post identity + duplicate collapse + supported claim spans + per-source usage + reviewer acceptance + bill. xAI cross-source search matched ledger; matched evidence/pricing registry verified 2026-08-27; unsupported fields fail closed as Unavailable. Dated first-party pricing/evidence registry, verified 2026-08-27. Module-local first-party sources: xAI search documentationxAI model pricing.

2. Collection parser page, section, and code-boundary integrity canary

Formula / scoring rule: Parser integrity = file/content/parser/chunk IDs + extracted span hashes/locations + boundary loss/duplication + searchable state + units + rollback + bill. xAI collection parser matched canary; matched evidence/pricing registry verified 2026-08-27; unsupported fields fail closed as Unavailable. Dated first-party pricing/evidence registry, verified 2026-08-27. Module-local first-party sources: xAI collections documentationxAI model pricing.

3. Realtime voice packet-loss, jitter, resampling, and duplicate-audio ledger

Formula / scoring rule: Audio settlement = packet/session IDs + committed/decoded duration + transcript continuity + duplicate effects + modality usage + concealment/reconnect + acceptance + charge. xAI live-audio-loss matched ledger; matched evidence/pricing registry verified 2026-08-27; unsupported fields fail closed as Unavailable. Dated first-party pricing/evidence registry, verified 2026-08-27. Module-local first-party sources: xAI Realtime documentationxAI model pricing.

1. X Search handle filter and account-identity transition ledger

Formula / scoring rule: Identity filter = submitted/effective handle controls + account/post/event IDs + creation/rename state + included results + claim support + usage + reviewer decision + latency + charge. xAI identity-transition matched ledger; matched evidence/pricing registry verified 2026-08-27; unsupported fields fail closed as Unavailable. Dated first-party pricing/evidence registry, verified 2026-08-27. Module-local first-party sources: xAI X Search documentationxAI model pricing.

2. Web Search redirect, canonical-collapse, paywall, and duplicate-source settlement canary

Formula / scoring rule: Canonical settlement = query/call/result/final-URL IDs + unique sources + supported claims + citation spans + search/model units + retry mutation + acceptance + marginal bill. xAI Web Search canonical matched canary; matched evidence/pricing registry verified 2026-08-27; unsupported fields fail closed as Unavailable. Dated first-party pricing/evidence registry, verified 2026-08-27. Module-local first-party sources: xAI Web Search documentationxAI model pricing.

3. Realtime voice function-call cancellation, barge-in, and exactly-once side-effect ledger

Formula / scoring rule: Voice side effects = session/response/call/result/event IDs + audio/transcript commit + cancellation acknowledgement + executed effects + modality/model usage + compensation + acceptance + invoice. xAI Realtime voice side-effect matched ledger; matched evidence/pricing registry verified 2026-08-27; unsupported fields fail closed as Unavailable. Dated first-party pricing/evidence registry, verified 2026-08-27. Module-local first-party sources: xAI Realtime voice documentationxAI model pricing.