Google API Pricing, Models & Rate Limits (2026)

1. Unit-safe add-on invoice ledger

Fixed multimodal-plus-grounding workload: 80K input, 8K output, 2 grounding calls, 1 code execution, 4 images, 60 seconds audio, 30 seconds video. Missing add-on rates remain Unavailable.

2. Free-tier exhaustion-to-paid crossover

Quota is not price: project/account limits and free-tier exhaustion are Unavailable unless Google documents them for the exact model and project.

3. AI Studio-to-Vertex production handoff canary

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. Context-cache storage-duration break-even

Formula: total = cache write + (reuses − 1) × cache read + storage duration + refreshed-prefix/miss spend. Values are shown only when compatible cache units and duration rates are sourced; ordinary input rates are not substituted.

2. Grounding-budget inverse planner

Inverse formula: maximum requests = floor((budget − sourced free allowance) ÷ (model-token cost + grounding-unit cost × search queries)). Grounding price, quota, free allowance, and quality impact remain separate evidence fields.

3. Online-versus-batch delivery ledger

Formula: online requests = total × (1 − deferred share); batch requests = total × deferred share. Submission, completion, failures, resubmission, deadline, and region/project eligibility are not inferred from a generic discount.

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. URL-context versus inline versus file-input parity

Formula / rule: parity=spans∧valid citations∧context fit∧reviewer Source: pricing registry verified 2026-08-26. Rate: Gemini 3.1 Pro, $2.0000 input/M + $12.0000 output/M.

2. Code-execution sandbox conformance audit

Formula / rule: accepted=model usage+declared tool units Source: pricing registry verified 2026-08-26. Rate: Gemini 3.1 Pro, $2.0000 input/M + $12.0000 output/M.

3. Online-versus-Batch multimodal result-parity ledger

Formula / rule: parity=request∧rubric∧finish/safety∧deadline Source: pricing registry verified 2026-08-26. Rate: Gemini 3.1 Pro, $2.0000 input/M + $12.0000 output/M.

1. Context-cache TTL-refresh cost ledger

Formula / rule: Illustrative with-cache-vs-without-cache bill = frozen-turn token bill at the Gemini 3.1 Pro registry rate, comparing a full-input baseline against an assumed reduced-input cache-hit case. Per-renewal storage charge and the actual returned cache-hit token savings require a sourced Vertex cache-storage rate and a matched renewed-session run, neither of which is present in the registry, so the figures are labelled illustrative-only. Source: pricing registry verified 2026-08-26.

2. Parallel function-calling determinism-and-error audit

Formula / rule: Base-request cost = (frozen prompt+schema tokens × input rate + response tokens × output rate)/1M at the Gemini 3.1 Pro registry rate. Call set, order, duplicate/omitted calls, and malformed-argument rate across repeated identical requests 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.

3. Vertex regional-failover fallback ledger

Formula / rule: Primary-region base bill = frozen-request-sequence token bill at the Gemini 3.1 Pro registry rate. Detected failure signal, fallback region selection, latency delta, retried-request duplication risk, cost delta versus the primary region, and SLA credit terms all require a declared regional-unavailability window and matched run, neither of which is present in the registry, so only the primary-region base bill below is reproducible. Source: pricing registry verified 2026-08-26.

1. Thinking-budget (reasoning-token) cost and latency tradeoff ledger

Formula / rule: Final-answer-only floor cost = (frozen prompt tokens × input rate + visible-output tokens × output rate)/1M at the Gemini 3.1 Pro registry rate, excluding the separately-billed thinking-token count. Per-tier thinking-token consumption and whether an unbounded/dynamic budget setting is priced identically to a fixed cap require a matched run at each documented tier, which is not present in the registry, so total cost including thinking tokens is Unavailable at every tier. Source: pricing registry verified 2026-08-26.

2. Vertex provisioned-throughput (committed-use) cost ledger

Formula / rule: Pay-as-you-go cost = frozen sustained-volume-workload token bill at the Gemini 3.1 Pro registry rate. A documented Vertex provisioned-throughput commitment rate is not present in the pricing registry, so the pay-as-you-go figures below are reproducible but no committed-use comparison or breakeven point can be computed. Source: pricing registry verified 2026-08-26.

3. Embeddings-endpoint cost-per-dimension ledger

Formula / rule: The pricing registry carries no dated Gemini embedding-model per-token rate or documented output-dimensionality-reduction billing rule, so cost-per-dimension is Unavailable rather than derived from the text-completion rate — distinct from the equivalent OpenAI embeddings ledger's own dimension-reduction behavior, which is separately Unavailable there too. Source: pricing registry verified 2026-08-26.

1. Implicit-caching hit-rate economics ledger

Formula / rule: No-cache-credit cost = frozen prompt-set token bill at full input rate at the Gemini 3.1 Pro registry rate, assuming zero implicit-cache credit. The documented implicit-cache hit-rate at each repeated-prefix-share level and the resulting discounted-token count require a matched repeated-request run, which is not present in the registry, so only the no-credit ceiling below is reproducible. Source: pricing registry verified 2026-08-26.

2. Batch-job-cancellation billing ledger

Formula / rule: Submitted-job cost = frozen batch-request-set token bill at the Gemini 3.1 Pro registry rate for the full submitted volume. Whether a documented mid-run cancellation bills only completed items, the full submitted batch, or a separate cancellation fee requires a sourced cancellation-billing policy and a matched cancelled-run record, neither of which is present in the registry, so only the full-submitted-volume figure below is reproducible. Source: pricing registry verified 2026-08-26.

3. Combined code-execution-plus-function-call cost ledger

Formula / rule: Base-request cost = (frozen prompt+tool-declaration tokens × input rate + response tokens × output rate)/1M at the Gemini 3.1 Pro registry rate, excluding any code-execution-sandbox surcharge. The isolated cost of the code-execution tool when combined in the same turn as a function call requires a matched combined-tool run, which is not present in the registry, so only the base-request cost below is reproducible. Source: pricing registry verified 2026-08-26.

1. Implicit-caching hit-rate economics ledger

Formula / rule: No-cache-credit cost = frozen prompt-set token bill at full input rate at the Gemini 3.1 Pro registry rate, assuming zero implicit-cache credit. The documented implicit-cache hit-rate at each repeated-prefix-share level and the resulting discounted-token count require a matched repeated-request run, which is not present in the registry, so only the no-credit ceiling below is reproducible. Source: pricing registry verified 2026-08-26.

2. Batch-job-cancellation billing ledger

Formula / rule: Submitted-job cost = frozen batch-request-set token bill at the Gemini 3.1 Pro registry rate for the full submitted volume. Whether a documented mid-run cancellation bills only completed items, the full submitted batch, or a separate cancellation fee requires a sourced cancellation-billing policy and a matched cancelled-run record, neither of which is present in the registry, so only the full-submitted-volume figure below is reproducible. Source: pricing registry verified 2026-08-26.

3. Combined code-execution-plus-function-call cost ledger

Formula / rule: Base-request cost = (frozen prompt+tool-declaration tokens × input rate + response tokens × output rate)/1M at the Gemini 3.1 Pro registry rate, excluding any code-execution-sandbox surcharge. The isolated cost of the code-execution tool when combined in the same turn as a function call requires a matched combined-tool run, which is not present in the registry, so only the base-request cost below is reproducible. Source: pricing registry verified 2026-08-26.

1. `countTokens` preflight versus `generateContent` returned-usage reconciliation

Formula / scoring rule: Preflight delta = returned modality/tool/cache usage − countTokens estimate; exact bill uses returned usage and the Gemini registry rate. A preflight estimate is never treated as an invoice. Source: pricing registry verified 2026-08-27.

2. Safety-blocked completion billing ledger

Formula / scoring rule: Base bill uses declared prompt tokens and observed output only when returned usage exists. Policy enforcement is not presumed free; blocked, visible, retry, and invoice fields must be matched. Source: pricing registry verified 2026-08-27.

3. Explicit cached-content early-delete proration audit

Formula / scoring rule: Final bill = creation token bill + sourced storage duration/delete/refund/recreation charges. Without a dated cache-storage/deletion rate, only the token baseline is shown. Source: pricing registry verified 2026-08-27.

1. URL Context fetch-and-token billing ledger

Formula / scoring rule: URL cost = fetched/extracted units × URL-context rate + model input/output bill + retry bill. Redirects, robots/paywalls, oversized documents, and citations are observed states, never inferred token counts. Source: pricing registry and dated evidence index verified 2026-08-27.

2. Batch API malformed-row atomicity and retry-subset invoice audit

Formula / scoring rule: Batch bill = Σ returned usage for accepted/completed rows + Σ retry-subset usage. Duplicate IDs, invalid modalities, and over-limit rows must be classified by the file/job validation scope; no all-file charge is assumed. Source: pricing registry and dated evidence index verified 2026-08-27.

3. Live API simultaneous audio-and-text output attribution ledger

Formula / scoring rule: Total bill = returned audio units × audio rate + returned text/transcription units × their rates + model token bill. Duplicated semantic content and interrupted turns are counted only from terminal usage. Source: pricing registry and dated evidence index verified 2026-08-27.

1. Cached-content cross-scope reuse-isolation canary

Formula / scoring rule: A cache hit is valid only when project, region, model, owner, and access-control scope match; bill uses returned hit/miss units, never an assumed reuse. Source: pricing registry and dated evidence index verified 2026-08-27.

2. System-instruction versus first-user-turn role accounting

Formula / scoring rule: Role delta = returned modality/cached/thinking usage in placement A − placement B; exact bill uses returned usage for each matched prompt. Source: pricing registry and dated evidence index verified 2026-08-27.

3. Multimodal function-response payload ledger

Formula / scoring rule: Total = function declaration + call arguments + returned media units + thought signature + final output; unsupported modality/rate remains Unavailable. Source: pricing registry and dated evidence index verified 2026-08-27.

1. candidateCount=1/2/4/8 acceptance and billing ledger

Formula / scoring rule: Total = shared prompt/media usage + Σ(candidate thinking/output) + repair; qualify each candidate only when its finish/safety state is accepted. Source: pricing registry and dated evidence index verified 2026-08-27.

2. Stop-sequence termination canary

Formula / scoring rule: Accepted cost = returned modality/thinking/output usage through the matched stop plus any continuation repair; suffix after stop is not inferred usage. Source: pricing registry and dated evidence index verified 2026-08-27.

3. Response-schema description/enum footprint ledger

Formula / scoring rule: Footprint delta = returned/preflight input tokens with schema − equivalent unconstrained prompt; bill uses returned usage and repair calls. Source: pricing registry and dated evidence index verified 2026-08-27.

1. text/plain versus application/json response-MIME bill ledger

Formula / scoring rule: Compare configuration support, schema footprint, returned thinking/output usage, parse state, repair, latency, and accepted cost. Source: pricing registry and dated evidence index verified 2026-08-27.

2. Media-resolution control canary

Formula / scoring rule: Compare decoded dimensions/duration, parameter acceptance, modality usage, context headroom, task score, retries, and bill per supported pair. Source: pricing registry and dated evidence index verified 2026-08-27.

3. Grounded-search domain and recency-filter conformance matrix

Formula / scoring rule: Conformance = accepted filter × eligible source/date fit × citation validity × joined search/model units; empty results require explicit recovery. Source: pricing registry and dated evidence index verified 2026-08-27.

1. Rendered-media normalization canary

Formula / scoring rule: Compare decoded dimensions/duration, modality usage, answer equivalence, context headroom, retries, and bill after changing only representation metadata. Source: pricing registry and dated evidence index verified 2026-08-27.

2. Protected-document and page-selection PDF ingestion ledger

Formula / scoring rule: Accepted PDF cost = extracted text/image units + model usage + repair/decryption transformation + storage state; unsupported protection remains narrowly Unavailable. Source: pricing registry and dated evidence index verified 2026-08-27.

3. Function-response payload boundary and truncation ledger

Formula / scoring rule: Accepted-result cost = schema + call + result input units + modality/thinking/output usage + repair/resend; silent truncation is not acceptance. Source: pricing registry and dated evidence index verified 2026-08-27.

1. Resumable File API interruption, checksum, duplicate, and orphan-state ledger

Formula / scoring rule: Final file state = session offset + accepted bytes + checksum + resume/retry scope + duplicate IDs + processing/expiry state + sourced units. Byte-storage charge remains Unavailable unless sourced. Source: pricing registry and dated evidence index verified 2026-08-27; provider registry: Google Gemini 2.0 Flash / File API registry rate verified 2026-08-27; test suite: Batch 30 Google File API lifecycle fixture/test suite (run and result recorded 2026-08-27).

2. Tuned-model failure/cancellation and inference-surcharge reconciler

Formula / scoring rule: Invoice = training units actually charged + retained state + base inference units + tuned inference surcharge − credits; failed and cancelled units cannot be treated as completed. Source: pricing registry and dated evidence index verified 2026-08-27; provider registry: Google Gemini 2.0 Flash / tuning-job registry rate verified 2026-08-27; test suite: Batch 30 Google tuning-job fixture/test suite (run and result recorded 2026-08-27).

3. Batch output-artifact retention and redownload ledger

Formula / scoring rule: Close = terminal row counts + result/error object state + download integrity + returned usage + storage/deletion state + retry subset + final bill; retention edge is probed, not assumed. Source: pricing registry and dated evidence index verified 2026-08-27; provider registry: Google Gemini 2.0 Flash / Batch API registry rate verified 2026-08-27; test suite: Batch 30 Google Batch-artifact fixture/test suite (run and result recorded 2026-08-27).

1. Live automatic-activity-detection, barge-in, and audio-transcript settlement ledger

Formula / scoring rule: Accepted turn = committed audio + transcript continuity + model usage + tool interruption state; discarded audio requires an event-backed state and is not presumed free. First-party registry: allaiask.com pricing and evidence registry, verified 2026-08-27. Provider/model source: Google Gemini Live pricing and event registry, verified 2026-08-27.

2. Dynamic-retrieval threshold marginal-yield ledger

Formula / scoring rule: Marginal yield = newly supported claims ÷ additional grounding queries; false citations, latency, reviewer acceptance, and thinking/output usage all join. First-party registry: allaiask.com pricing and evidence registry, verified 2026-08-27. Provider/model source: Google Gemini grounding and pricing registry, verified 2026-08-27.

3. URL-context redirect, authorization, robots, expiry, timeout, and MIME failure-debit audit

Formula / scoring rule: Close = fetch state + finish/prompt feedback + visible output + returned modality/cache/thinking/output usage + retry transport + invoice; failure is never zero by assumption. First-party registry: allaiask.com pricing and evidence registry, verified 2026-08-27. Provider/model source: Google Gemini URL context and pricing registry, verified 2026-08-27.

1. Explicit-cache create/get/update/expire/delete propagation ledger

Formula / scoring rule: Cache close = state transition + reference hit/miss + mutation/deletion propagation + create/storage units + final bill. Google Gemini cache-object and pricing evidence registry. Dated registry and evidence index, verified 2026-08-27.

2. Safety-block and partial-candidate settlement matrix

Formula / scoring rule: Accepted settlement = returned input/cache/thinking/output usage plus reviewer classification; a blocked candidate is not zero cost. Google Gemini safety settings and pricing evidence registry. Dated registry and evidence index, verified 2026-08-27.

3. Grounding-plus-function-calling composition ledger

Formula / scoring rule: Marginal bill = grounding/search/model units + function calls/results + retries; unsupported simultaneous combinations stay Unavailable. Google Gemini grounding, function-calling, and pricing evidence registry. Dated registry and evidence index, verified 2026-08-27.

1. Per-part `media_resolution` acceptance and token-attribution ledger

Formula / scoring rule: Media bill = returned per-modality/input/output usage at the accepted resolution; decoded dimensions, pages, and duration remain visible inputs. First-party pricing/evidence registry: Google Gemini media resolution and token/pricing evidence, verified 2026-08-27; unsupported units or credits remain Unavailable.. Dated registry and evidence index, verified 2026-08-27. First-party sources: Google Gemini media resolution documentationGoogle Gemini pricing.

2. Thought-summary requested-versus-omitted exposure and settlement canary

Formula / scoring rule: Charge = disclosed returned usage, including final output; a thought summary is not the full reasoning trace and hidden units are not guessed. First-party pricing/evidence registry: Google Gemini thinking controls, summaries, 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: Google Gemini thinking documentationGoogle Gemini pricing.

3. `systemInstruction` versus cached-content instruction-precedence ledger

Formula / scoring rule: Precedence result = effective instruction adherence + cache version/invalidation + regression outcome; cached input is not assumed free. First-party pricing/evidence registry: Google Gemini systemInstruction, cached content, 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: Google Gemini context caching documentationGoogle Gemini pricing.

1. Google Maps grounding place-identity, route, and marginal-cost ledger

Formula / scoring rule: Grounded acceptance = place/route identity + citation support + qualifier fidelity + reviewer result; Maps units use returned usage only. Google Maps grounding and Gemini 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: Google Maps groundingGoogle Gemini pricing.

2. Live API ephemeral-token mint, scope, expiry, reuse, and revocation canary

Formula / scoring rule: Session acceptance = token scope/expiry/revocation + handshake/session identity + retained context/audio + returned usage and charge. Google Gemini Live ephemeral authentication 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: Google Live API ephemeral tokensGoogle Gemini pricing.

3. OpenAI-compatible endpoint versus native Gemini endpoint invoice-parity ledger

Formula / scoring rule: Parity = jointly supported wire fields + effective config/model + usage + semantic acceptance + invoice match; SDK shape alone is insufficient. Google Gemini native/OpenAI-compatible endpoint 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: Google Gemini OpenAI compatibilityGoogle Gemini pricing.

1. SDK automatic-function-calling versus manual loop ledger

Formula / scoring rule: Hidden-round-trip acceptance = effective request/model IDs + hidden call count + arguments/results + per-round usage + semantic acceptance + bill. Google automatic-function-calling 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: Google function calling documentationGoogle Gemini pricing.

2. Live proactive GoAway and session-expiry continuation canary

Formula / scoring rule: Continuation acceptance = connection/session IDs + warning/close events + resumption handle + retained state + duplicate suppression + usage + charge. Google Live GoAway and expiry 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: Google Live API documentationGoogle Gemini pricing.

3. Vertex request-label propagation and cost-export attribution ledger

Formula / scoring rule: Attribution acceptance = submitted/effective labels + request/job IDs + usage/export rows + lag/collision state + reconciliation total. Google Vertex request-label export 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: Google Vertex request labels documentationGoogle Gemini pricing.

1. Stateful Interactions versus stateless GenerateContent ledger

Formula / scoring rule: Stateful result = interaction/predecessor/request/model IDs + retained/resend content + tool/thought state + usage + branch/deletion behavior + migration fallback + bill. Google stateful interaction 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: Google Interactions API documentationGoogle Gemini model pricing.

2. Dynamic Google Search retrieval admission and threshold-settlement canary

Formula / scoring rule: Grounding decision = submitted/effective mode + threshold + retrieval score/queries/sources + no-search reason + supported claims + model/grounding usage + marginal charge. Google dynamic Search grounding 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: Google Grounding documentationGoogle Gemini model pricing.

3. Explicit cached-content cross-project and cross-principal isolation ledger

Formula / scoring rule: Isolation result = project/location/cache/model identity + IAM decision/propagation + cached/uncached usage + leakage/denial + recreation + equivalence + invoice. Google cached-content principal isolation 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: Google context caching documentationGoogle Gemini model pricing.

1. Simultaneous response MIME/schema and function-tool arbitration ledger

Formula / scoring rule: Arbitration = submitted/effective configuration + selected response/tool path + schema/argument validity + finish + usage + repair + semantic acceptance + bill. Google response-tool arbitration 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: Google function calling documentationGoogle Gemini model pricing.

2. Files resumable-upload chunk, checksum, retry, and duplicate-finalize canary

Formula / scoring rule: Upload settlement = session/file/content hashes + accepted byte ranges + checksum + processing state + duplicate identity + storage/input usage + retry + invoice. Google resumable-upload 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: Google Files API documentationGoogle Gemini model pricing.

3. Live context-window-compression trigger and state-fidelity ledger

Formula / scoring rule: Compression fidelity = session/event IDs + effective trigger + tokens before/after + retained instructions/transcript/tool state + usage + dropped facts + recovery + charge. Google Live compression 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: Google Live API documentationGoogle Gemini model pricing.

1. URL Context redirect, canonical, fragment, encoding, and mixed-fetch atomicity ledger

Formula / scoring rule: URL settlement = requested/final URL and retrieval metadata + cited spans + cache/modality/input/thinking/output usage + partial answer + retry subset + acceptance + bill. Google URL Context 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: Google URL Context documentationGoogle Gemini model pricing.

2. Google Search grounding image-result provenance and settlement canary

Formula / scoring rule: Grounded-image result = tool configuration + search/query/result/image/source IDs + decoded media + claim-to-pixel/text support + usage + acceptance + retry + marginal charge. Google grounded-image 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: Google Google Search grounding documentationGoogle Gemini model pricing.

3. Code-execution stdout, stderr, exit, timeout, generated-file, and continuation ledger

Formula / scoring rule: Execution artifact = code/outcome/artifact IDs + container state + file hashes + cache/input/thinking/output/tool usage + restart/replay + checker + retention/download + invoice. Google code-execution artifact 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: Google code execution documentationGoogle Gemini model pricing.