# 安価API・無料枠活用大全

CURRENT_STATUS: CURRENT
OBSERVATION: SPEC_ONLY
FRESHNESS: CURRENT
PAGE_TYPE: reference
CANONICAL_URL: /tools/ea-008-api
CLAIM_IDS: none
EVIDENCE_IDS: EVD-EA-008-URL-B88C4950BBF3, EVD-EA-008-URL-2F2B6946DECD, EVD-EA-008-URL-AB27080818D0, EVD-EA-008-URL-0BE48C3457DF, EVD-EA-008-URL-9263769683C1, EVD-EA-008-URL-C03A6DE01B20, EVD-EA-008-URL-6F9CE73E67E4, EVD-EA-008-URL-F51D2DF41C34, EVD-EA-008-URL-77EEE737D65E, EVD-EA-008-URL-00EAF917AB11, EVD-EA-008-URL-2902590D1C1A, EVD-EA-008-URL-50846CD1FF36, EA-008-CURRENT-APPROVAL, EA-008-CURRENT-BODY-READBACK
UNKNOWN_IDS: EA-008-UNKNOWN-01
OPS_IDS: none

## Summary
Compare low-cost API use with an as-of ledger that separates list price, free allowance, trial credit, quota/rate limit, cache/batch/tool charges, region, data-use terms, and migration-to-paid conditions. Re-read the provider's current billing page before procurement or production rollout.

## Current judgment
- Current provider pricing still demonstrates why “free” and “cheap” must be modeled as dated commercial/technical contracts rather than permanent model properties.
- OpenAI's current API pricing separates model token rates and can add regional-processing uplifts for eligible data-residency endpoints.
- Gemini's current pricing distinguishes Free Tier, Standard, Batch/Flex/Priority, tool charges, and shows data-use treatment can differ between free and paid tiers for some models.
- Anthropic maintains model/platform pricing with model, scope, cache, batch, and regional dimensions.
- Token price alone remains insufficient: quota, rate limit, tool charges, cache/storage, data-use terms, region, model availability, trial/credit expiration, and paid-tier transition require an as-of date.

## Scope
安価API・無料枠活用大全

## Confirmed
- Current provider pricing still demonstrates why “free” and “cheap” must be modeled as dated commercial/technical contracts rather than permanent model properties.
- OpenAI's current API pricing separates model token rates and can add regional-processing uplifts for eligible data-residency endpoints.
- Gemini's current pricing distinguishes Free Tier, Standard, Batch/Flex/Priority, tool charges, and shows data-use treatment can differ between free and paid tiers for some models.
- Anthropic maintains model/platform pricing with model, scope, cache, batch, and regional dimensions.
- Token price alone remains insufficient: quota, rate limit, tool charges, cache/storage, data-use terms, region, model availability, trial/credit expiration, and paid-tier transition require an as-of date.

## UNKNOWN
- EA-008-UNKNOWN-01: - negotiated enterprise discounts and organization-specific credits;
- account/region-specific quota assignments;
- promotional programs not expressed in stable public pricing;
- future model substitutions and automatic tier migrations.

## Evidence
- EVD-EA-008-URL-B88C4950BBF3 (RESEARCH): | E08-01 | official pricing | Gemini Developer API pricing — Google | — https://ai.google.dev/gemini-api/docs/pricing
- EVD-EA-008-URL-2F2B6946DECD (RESEARCH): | E08-02 | official billing | Billing — Gemini API — Google | — https://ai.google.dev/gemini-api/docs/billing
- EVD-EA-008-URL-AB27080818D0 (RESEARCH): | E08-03 | official pricing | Pricing and Billing — Hugging Face Inference Providers | — https://huggingface.co/docs/inference-providers/pricing
- EVD-EA-008-URL-0BE48C3457DF (RESEARCH): | E08-04 | official pricing | Pricing — Cloudflare Workers AI | — https://developers.cloudflare.com/workers-ai/platform/pricing/
- EVD-EA-008-URL-9263769683C1 (RESEARCH): | E08-05 | official changelog | Select models now require the Workers Paid plan — Cloudflare | — https://developers.cloudflare.com/changelog/post/2026-07-28-models-require-workers-paid/
- EVD-EA-008-URL-C03A6DE01B20 (RESEARCH): | E08-06 | official limits | Rate Limits — GroqDocs | — https://console.groq.com/docs/rate-limits
- EVD-EA-008-URL-6F9CE73E67E4 (RESEARCH): | E08-07 | official billing | Billing FAQs — GroqDocs | — https://console.groq.com/docs/billing-faqs
- EVD-EA-008-URL-F51D2DF41C34 (RESEARCH): | E08-08 | official API ops/pricing | Batch API FAQ — OpenAI | — https://help.openai.com/en/articles/9197833-batch-api-faq
- EVD-EA-008-URL-77EEE737D65E (RESEARCH): | E08-09 | official model pricing | GPT-5.5 Model — OpenAI API | — https://developers.openai.com/api/docs/models/gpt-5.5
- EVD-EA-008-URL-00EAF917AB11 (RESEARCH): https://developers.openai.com/api/docs/pricing — https://developers.openai.com/api/docs/pricing
- EVD-EA-008-URL-2902590D1C1A (RESEARCH): https://ai.google.dev/gemini-api/docs/billing/ — https://ai.google.dev/gemini-api/docs/billing/
- EVD-EA-008-URL-50846CD1FF36 (RESEARCH): https://platform.claude.com/docs/en/about-claude/pricing — https://platform.claude.com/docs/en/about-claude/pricing
- EA-008-CURRENT-APPROVAL (RESEARCH): 現行本文の監査・revision対応
- EA-008-CURRENT-BODY-READBACK (IMPL): 現行本文の収録照合

## Failure / limits
Do not publish a timeless “cheapest provider” ranking from a dated snapshot.
## Failure / limits
Free-tier availability is not equivalent to a contractual production SLA or stable quota.
## Failure / limits
Promotional credits, free requests, batch discounts, caching discounts, and paid-plan included allowances are different economic mechanisms.
## Failure / limits
Privacy/data-use treatment can differ by tier or product path; zero price must not be treated as zero governance cost.
## Failure / limits
Tool/grounding/search and regional-processing charges may dominate token-only estimates.
## Practice
Current provider pricing still demonstrates why “free” and “cheap” must be modeled as dated commercial/technical contracts rather than permanent model properties.
## Practice
OpenAI's current API pricing separates model token rates and can add regional-processing uplifts for eligible data-residency endpoints.
## Practice
Gemini's current pricing distinguishes Free Tier, Standard, Batch/Flex/Priority, tool charges, and shows data-use treatment can differ between free and paid tiers for some models.
## Practice
Anthropic maintains model/platform pricing with model, scope, cache, batch, and regional dimensions.
## Practice
Token price alone remains insufficient: quota, rate limit, tool charges, cache/storage, data-use terms, region, model availability, trial/credit expiration, and paid-tier transition require an as-of date.
## EA-008 安価API・無料枠活用大全
### WHY_THIS_MATTERS_TO_UPDATE
公開後アップデートで「安いAPI」「無料枠」を製品名と単価だけで並べると、恒常free allowance、試用credit、rate-limitだけのfree plan、paid plan内の無償割当、batch/cache等の割引を混同する。さらに、無料利用と有料利用でdata-use、model availability、quota scope、rate limit、billing移行条件が変わるproviderもある。したがって本項は価格表ではなく、低コスト利用の運用境界を比較可能にする辞典として設計する必要がある。


### SCOPE_ROUTING
Enterprise AI Core / provider pricing & availability / organization cost governance / implementation / operability / privacy / benchmark-adjacent cost evaluation


### CLAIM_OR_RESEARCH_QUESTION
1. 「無料」は単一カテゴリではなく、free tier、recurring credit、daily allowance、rate-limited plan、paid-plan included allowance等へ分解できるか。
2. free→paidの遷移時に、billing method、prepay、quota/rate limit、data-use、model availabilityはどう変わるか。
3. 単価以外のcost control（Batch、cached input、quota/budget guard等）をfree tierと分離して記録すべきか。
4. 将来の辞典でprovider横断比較するための最小フィールドは何か。


### EVIDENCE TABLE
| ID | SOURCE_CLASS | exact title / author-org | canonical URL | publication/update/access | version / locator | LABEL | RELATION |
|---|---|---|---|---|---|---|---|
| E08-01 | official pricing | Gemini Developer API pricing — Google | https://ai.google.dev/gemini-api/docs/pricing | accessed 2026-09-11; page published 2026-09-08 | Free / Paid / Enterprise; Gemini 3.8 Flash pricing tables | VERIFIED_SPEC / PRODUCT_VOLATILE | support, limit |
| E08-02 | official billing | Billing — Gemini API — Google | https://ai.google.dev/gemini-api/docs/billing | accessed 2026-09-11; page published 2026-09-04 | Free, Tier 1-3; prepaid minimum; billing-account-level caps | VERIFIED_SPEC / PRODUCT_VOLATILE | support |
| E08-03 | official pricing | Pricing and Billing — Hugging Face Inference Providers | https://huggingface.co/docs/inference-providers/pricing | accessed 2026-09-11 | monthly credits; routed vs custom provider key | VERIFIED_SPEC / PRODUCT_VOLATILE | support, limit |
| E08-04 | official pricing | Pricing — Cloudflare Workers AI | https://developers.cloudflare.com/workers-ai/platform/pricing/ | updated 2026-08-28; accessed 2026-09-11 | 10,000 Neurons/day; Free/Paid behavior | VERIFIED_SPEC / PRODUCT_VOLATILE | support, limit |
| E08-05 | official changelog | Select models now require the Workers Paid plan — Cloudflare | https://developers.cloudflare.com/changelog/post/2026-07-28-models-require-workers-paid/ | 2026-07-28; accessed 2026-09-11 | model-level paid gate / 403 | VERIFIED_SPEC / PRODUCT_VOLATILE | counter, limit |
| E08-06 | official limits | Rate Limits — GroqDocs | https://console.groq.com/docs/rate-limits | accessed 2026-09-11 | Free Plan Limits; RPM/RPD/TPM/TPD; rate-limit headers | VERIFIED_SPEC / PRODUCT_VOLATILE | support |
| E08-07 | official billing | Billing FAQs — GroqDocs | https://console.groq.com/docs/billing-faqs | accessed 2026-09-11 | Free→Developer; payment method; progressive billing thresholds | VERIFIED_SPEC / PRODUCT_VOLATILE | support |
| E08-08 | official API ops/pricing | Batch API FAQ — OpenAI | https://help.openai.com/en/articles/9197833-batch-api-faq | updated Aug 2026; accessed 2026-09-11 | asynchronous Batch; 50% discount vs synchronous API | VERIFIED_SPEC / PRODUCT_VOLATILE | context, support |
| E08-09 | official model pricing | GPT-5.5 Model — OpenAI API | https://developers.openai.com/api/docs/models/gpt-5.5 | accessed 2026-09-11 | per-1M token standard/cached pricing; Batch-supported model page | VERIFIED_SPEC / PRODUCT_VOLATILE | context, support |


### FINDINGS


#### 1. 「無料枠」は最低でも5種に分ける
- **FREE_TIER_PRICE_ZERO**: 対象model/input/outputが0円だが、model・rate・tool等に制限がある。Google Gemini Developer APIが典型で、Free tierは一部modelへ限定され、Paidはhigher rate limitsやadvanced featuresへ広がる。Free tierの入力・出力が0円でも、すべてのmodel/tool/modeがfreeではない。
- **RECURRING_CREDIT**: 毎月一定額のcreditが付与され、超過後は購入/PAYGへ移る。Hugging Face Inference ProvidersはFree user $0.10/month（subject to change）、PRO $2、Team/Enterprise $2/seatのmonthly creditsを明示する。これは「API単価0円」ではなくcredit消費型。
- **DAILY_INCLUDED_ALLOWANCE**: 一定量まで無料でresetする。Cloudflare Workers AIは10,000 Neurons/dayをFree/Paid双方に含み、00:00 UTCにresetする。Free planは超過分をそのまま従量課金せず、Paid planへのupgradeが必要。
- **RATE_LIMITED_FREE_PLAN**: 金額quotaよりRPM/RPD/TPM/TPDが実効上限になる。Groq Free Planはmodel別の複数rate limitを持ち、429とretry-after/remaining/reset headersを返す。
- **PAID_PLAN_INCLUDED_ALLOWANCE**: 有料planにもfree allocationが含まれる場合があり、free accountと同義ではない。Cloudflare Workers Paidの10,000 Neurons/dayやHF paid subscriptionのmonthly creditsが該当する。


したがって「無料／有料」の二値列は採用しない。


#### 2. rate limit と spend cap は別物
GroqのFree PlanはRPM/RPD/TPM/TPDで制御され、CloudflareはNeuron daily allocationとplan gateを持つ。429を受けることは「予算を守れた」の証明ではなく、逆にPAYG providerのbudgetを設定してもrate capacityは保証されない。辞典では `RATE_LIMIT` と `SPEND_CAP/BUDGET` を別列にする。


#### 3. free→paid遷移は単なる上限解除ではない
Google Gemini APIではbilling account連携とprepayによってPaid Tierへ移り、tier/capはbilling-account levelで決まる。さらにFreeは「content used to improve products: Yes」、PaidはNoと公式pricingが分離している。つまり価格遷移はprivacy/data-use境界の変更でもある。


GroqはFree→Developerでpayment methodが必要となり、PAYGとhigher token limits/supportへ移る。新規Developer利用者にはprogressive billing thresholdsがある。したがって「card登録の有無」「prepay/PAYG」「請求発火条件」をavailability/cost governanceの一部として記録する。


#### 4. free allocationがあってもmodel availabilityは別gate
Cloudflareは2026-07-28に一部resource-intensive modelをWorkers Freeから外し、Free planでは該当modelに403を返す仕様へ変更した。一方、10,000 Neurons/dayというfree allocation自体は維持される。よって `FREE_QUOTA_EXISTS=true` から `TARGET_MODEL_AVAILABLE=true` を推論してはならない。


Googleでもmodel/mode/toolごとにFree Tier availabilityが異なり、Batch/FlexやComputer Use等でFree Tier `Not available` が存在する。model IDだけでなくmode/tool単位のgateを持つ必要がある。


#### 5. aggregator経由とprovider直契約で同じmodelでもbilling境界が変わる
Hugging Face Inference Providersでは、HF routed requestならmonthly creditsが適用される一方、Custom Provider Keyではcreditsは適用されずproviderから直接請求される。したがってcost provenanceには `BILLING_ROUTE` が必要。model名だけでは請求主体・credit適用可否を判定できない。


#### 6. 「安価化」はfree tierとは別の最適化面
OpenAI Batch APIは同期API比50% discountを明示し、現行model pagesにはcached inputの別単価もある。これはfree allowanceではなくworkload placement/cachingによる単価低減である。将来章では少なくとも以下を分離する。
- free allowance / recurring credits
- smaller/cheaper model selection
- batch/offline discount
- cached-input discount
- tool-specific charges
- long-context surcharge等のusage-shape依存料金
- retry/replayによる増幅


「最安model」を選ぶだけでは総費用最小化にならない。


### LIMITS / COUNTEREVIDENCE / FAILURE ENVELOPE
- PRICE_VOLATILITY: 単価・free quota・model availability・promotional pricingは高頻度で変わる。固定額は必ずas-of付きで扱う。
- FREE_IS_NOT_SLA: free accessの存在はproduction SLA、capacity、support、latency保証を意味しない。
- QUOTA_SCOPE_MISMATCH: account/project/billing account/organization/modelごとにquota scopeが異なる。単純合算禁止。
- TOOL_COST_ESCAPE: tokenが無料/安価でもSearch/Maps/tool call/storage等が別meterになる場合がある。
- DATA_USE_BOUNDARY: 無料と有料でdata-use条件が違うproviderがある。コストだけでfree tierを推奨しない。
- MODEL_GATE: 同provider内でもfree対象外model/modeがある。
- RETRY_AMPLIFICATION: 429/5xxへの無制御retryはtoken/request課金とrate pressureを増幅し得る。429を「無料枠終了」と一律解釈しない。
- CROSS_PROVIDER_QUALITY: 1M tokens単価だけでは品質、latency、context、tool capability、structured output、failure rateが違うため実仕事単位の費用比較にならない。


### AVAILABILITY_GATES
比較時に最低限別列化する:
`ACCOUNT_TYPE / BILLING_REQUIRED / PAYMENT_METHOD / PREPAY_REQUIRED / FREE_CLASS / MODEL_ID / MODE / TOOL / REGION / ORG_OR_PROJECT_SCOPE / RATE_TIER / DATA_USE_TIER / SUPPORT_OR_SLA_TIER`。


### IMPLEMENTATION_STATE
- Google Gemini: Free/Paid/Enterpriseの明示的tieringとmodel/mode/tool別pricingが現行official docsに実装済み。
- Hugging Face: monthly credit + routed billing + custom provider keyの二経路が実装済み。
- Cloudflare Workers AI: daily free Neuron allocation + Paid overage + model-level paid gateが実装済み。
- GroqCloud: Free plan model別rate limits + Developer PAYG移行が実装済み。
- OpenAI: Batch/cached-input等のcost optimization mechanismは実装済み。今回のpacketではOpenAIについて「恒常free API tier」の存在を主張しない。


### BENCHMARK_OR_EVAL
価格表比較だけでは不十分。将来の実用比較は同一bounded workloadで `successful task / total input tokens / cached tokens / output tokens / tool calls / retries / wall-clock / failure count / billed amount` を記録し、**cost per accepted task** を算出する構造が望ましい（PROPOSAL）。model品質benchmarksを単価表へ直結させず、実際の業務受入条件を満たしたrunだけを分母・分子に使う。


### SECURITY / PRIVACY / AUDIT / COST / OPERABILITY
- **PRIVACY**: Google Gemini Free/Paidでproduct-improvement data-use条件が異なるため、free-tier採用判断にはdata classification gateが必要。
- **COST**: monthly/day credits、prepay、PAYG、billing threshold、tool fee、batch/cache discountを別ledgerにする。
- **AUDIT**: provider console usageだけに依存せず、内部でprovider/model/mode/billing-route/request count/token/tool/retry/cost estimateを記録する設計が望ましい（PROPOSAL）。
- **OPERABILITY**: response headersでremaining/resetが得られる場合はbackoff/schedulingへ利用できるが、providerごとに意味が違うため共通値へ無理に変換しない。
- **SECURITY**: 「無料だから検証用データを入れてよい」としない。privacy/data-use条件を先に評価する。


### PROPOSED COMPARISON MATRIX
将来Workがprovider横断表を作る際の最小列（PROPOSAL）:
`AS_OF | PROVIDER | SERVICE | MODEL/MODE | FREE_CLASS | BILLING_ROUTE | BILLING_REQUIRED | PAYMENT/PREPAY | UNIT | INCLUDED_QUOTA | QUOTA_SCOPE | RESET/EXPIRY | RATE_LIMIT | OVERAGE_BEHAVIOR | MODEL/TOOL_GATE | DATA_USE | SLA/SUPPORT | REGION/ORG_GATE | BATCH/CACHE_DISCOUNT | TOOL/ADJACENT_COST | LAST_VERIFIED`


### SAFE_WORDING_FOR_FUTURE_BOOK
「AI APIの『無料』は一種類ではない。恒常free tier、毎月credit、日次allocation、rate-limit型free plan、有料plan内の無償枠は、超過時挙動も請求主体も異なる。さらに無料／有料の境界がmodel availability、rate limit、data-use条件を変える場合がある。運用では単価ではなく、quota scope、reset、billing route、availability gate、privacy、retryを同時に管理する。」


### UPDATE_WORK_HANDOFF
1. 公開直前に全価格・quotaを再取得し、金額値は`PRODUCT_VOLATILE`としてas-of固定する。
2. 各providerについてpricing pageだけでなくrate-limit/billing/data-use/termsを一組で確認する。
3. `PROPOSED COMPARISON MATRIX`を正規表としてprovider別行を埋める。複数mode/toolは別行。
4. 後続追加候補: AWS Bedrock / Azure AI Foundry / GitHub Models / Together / Fireworks / Cerebras等。追加時も公式pricing+limits+data-useを同時取得し、二次比較サイトだけで順位付けしない。
5. 「最安ランキング」は作らず、用途profile（prototype、batch、high-throughput、private data、tool-heavy等）ごとにcost envelopeを算出する。
6. TEST_PROFILE（未実施）: synthetic/public payloadのみを使い、zero-spendまたは明示的hard cap環境でdocumented quota境界の直前まで実行し、429/403、retry-after、remaining/reset header、usage meter反映を観測する。billing有効化・契約変更・課金発生操作は本調査では実施しない。


### OPEN_UNKNOWN / STOP_CONDITION
- UNKNOWN: provider横断で統一されたfree-tier taxonomy/metadata standardは今回確認していない。
- UNKNOWN: 各providerのproduction SLA/supportとfree tierの厳密な対応は本packetでは網羅しない。
- UNKNOWN: 地域・tax・currency・enterprise negotiated pricingはorganization-specificで公開価格から閉じない。
- STOP_CONDITION: free/credit/allocation/rate/billing/data-use/model gate/batch-cacheを分離し、複数providerの一次資料で反例を確保したためEA-008は`BOUNDED_CLOSE`。金額は公開直前再検証を必須とする。


### RUN STATUS
EA-008: BOUNDED_CLOSE

## Currentness findings
- Current provider pricing still demonstrates why “free” and “cheap” must be modeled as dated commercial/technical contracts rather than permanent model properties.
- OpenAI's current API pricing separates model token rates and can add regional-processing uplifts for eligible data-residency endpoints.
- Gemini's current pricing distinguishes Free Tier, Standard, Batch/Flex/Priority, tool charges, and shows data-use treatment can differ between free and paid tiers for some models.
- Anthropic maintains model/platform pricing with model, scope, cache, batch, and regional dimensions.
- Token price alone remains insufficient: quota, rate limit, tool charges, cache/storage, data-use terms, region, model availability, trial/credit expiration, and paid-tier transition require an as-of date.

## Current primary evidence
- https://developers.openai.com/api/docs/pricing
- https://ai.google.dev/gemini-api/docs/pricing
- https://ai.google.dev/gemini-api/docs/billing/
- https://platform.claude.com/docs/en/about-claude/pricing

## Counterevidence / failure envelope retained
- Do not publish a timeless “cheapest provider” ranking from a dated snapshot.
- Free-tier availability is not equivalent to a contractual production SLA or stable quota.
- Promotional credits, free requests, batch discounts, caching discounts, and paid-plan included allowances are different economic mechanisms.
- Privacy/data-use treatment can differ by tier or product path; zero price must not be treated as zero governance cost.
- Tool/grounding/search and regional-processing charges may dominate token-only estimates.

## SAFE_WORDING
Compare low-cost API use with an as-of ledger that separates list price, free allowance, trial credit, quota/rate limit, cache/batch/tool charges, region, data-use terms, and migration-to-paid conditions. Re-read the provider's current billing page before procurement or production rollout.

## OPEN_UNKNOWN
- negotiated enterprise discounts and organization-specific credits;
- account/region-specific quota assignments;
- promotional programs not expressed in stable public pricing;
- future model substitutions and automatic tier migrations.

## Research Gate decision
`PASS_FOR_BOUNDED_BASE_REVISION`.

## Candidate implementation overlay
- target: `content/staging/EA-008.md`
- base repository commit: `97d3d104ae0b16a95e7dffddc3a8cb327795a796`
- path precondition: `ABSENT`
- candidate: `V21-CANDIDATE-EA008-BASE-20261011-C1`
- READY: NOT_CREATED
- AUDIT_PASS: NOT_CLAIMED
- GIT_WRITE: NOT_PERFORMED

NEXT: /agent/start.json
STOP: This is a knowledge projection, not an execution order.