2027 planning guide · Evidence checked 20 November 2026

Flat-rate AI API subscription: unlimited, with no monthly token cap

A flat-rate subscription AI model API with unlimited use and no monthly cap prices API access as one monthly fee and replaces the token meter with published fair-use limits. Plugsky Builder ($60/mo) and Scale ($120/mo) meet that definition; this page shows which other providers do, and where the label hides a request cap.

Short answer. A flat-rate AI API subscription charges one monthly price instead of metering tokens. Plugsky Builder ($60/mo) and Scale ($120/mo) have no monthly token cap and no per-token overage; fair-use concurrency (10 or 16 requests) and RPM limits (600 or 1,200) still apply. Unlimited never means infinite capacity.

What a flat-rate AI API subscription is

A flat-rate API is a commercial model, not a performance guarantee. You pay a fixed monthly fee, the provider does not meter input and output tokens against a monthly quota, and the invoice does not change when a prompt gets longer. Capacity is controlled by concurrency, requests per minute, context ceilings and a fair-use policy instead of a token meter.

That is different from four pricing models that are often marketed with the same words. The table separates them, because “flat” and “unlimited” are frequently used for plans that still count something.

Four API pricing models and what actually limits usage
ModelWhat you pay forBill predictabilityWhat usually caps youBest fit
Flat-rate subscriptionOne monthly price for included usageHigh: the plan price is the billConcurrency, RPM, context, fair useSteady production with long prompts or agents
Prepaid creditsA balance drawn down at per-token ratesMedium: fixed only if you cap the balanceThe balance, plus provider rate limitsSpiky, seasonal or low-volume work
Per-token billingInput and output tokens at published ratesLow: the invoice follows usageBudget, spend limits, rate limitsVariable volume or one specific model
Seat-based UI planOne price per human user per monthHigh for seats, not for API callsMessages or usage per seatPeople using a chat app, not a product API

A seat-based plan is not an API plan. If a tool charges per editor seat and limits messages per seat, putting it behind a product creates both a licensing problem and a capacity problem. For API work, compare flat-rate, credits and per-token only.

How we evaluated this

We compared providers on six published criteria: billing model, monthly token cap, request cap, concurrency or fair use, model breadth and context, and API compatibility. We also recorded regions because “unlimited” is useless if the data cannot be processed in the required jurisdiction. Only official provider pages counted; listicles were used to find candidates, not to set prices. No weights were applied and no composite score was calculated. Every row was checked on 20 November 2026, and “not published” means the provider does not state the number on that page, not that the limit is absent.

Flat-rate API providers compared

The eight providers below sell, or have recently sold, something described as a flat monthly API.

Flat-rate API plans: token caps, overage, request caps, concurrency, model breadth, context, regions and compatibility
ProviderBilling modelMonthly token capOverageRequest capConcurrency / fair useModel breadthContextRegionsAPI compatibility
PlugskyFlat monthly; annual 20% offNone on Builder/Scale; daily caps on Hobby/StarterNone on Builder/Scale; Hobby and Starter stop at daily capsNone on Builder/Scale; daily request caps below10 / 16 concurrent; 600 / 1,200 RPM39 public model IDs256K / 512K plan input; 32K outputUS, EU, GCC, APACOpenAI-compatible
CheapestInferenceFlat monthly with 8-hour blocksNone during an active blockNone; access ends with the blockNone stated; requests declined outside blocksOne request at a time2 models per pool, 6 across three poolsNot publishedTime blocks for APAC, Europe, Americas; not residencyOpenAI and Anthropic SDKs
AI2YouFlat per model, $50/moNoneNoneNot publishedNot published1 live model; 3 announced262K on the live model“Multi-region”; not specifiedOpenAI and Anthropic
UNLIFlat monthly plus request quotaNone per requestNone; request quota applies100–200,000 requests/month by plan20–200 RPMMulti-vendor; count not publishedNot publishedNot publishedOpenAI and Anthropic
APIClawFlat monthly plus daily request quotaNone per requestNone; request quota applies500–8,000 weighted requests/dayNot published11 model IDs listedUp to 1MNot publishedOpenAI-compatible
FeatherlessFlat $25 chat plan; API is metered creditsNone on chat; metered on APIChat none; API metered per tokenNot published4 chat units; 100 developer units40,000+ open-weight models32K chat; up to 256K developerNot publishedOpenAI-compatible
Awan LLMFlat monthly plus daily request capsNoneNone; request quota applies200 small / 10 medium / 10 large per day on Free; unlimited on Max20–200 RPM; parallel requests on Pro and MaxSmall, medium and large model tiersNot publishedNot publishedNot stated on the pricing page
GiottoPer-token API; no flat plan on the API pageNot applicablePer-token billingNot publishedNot published2 modelsNot publishedEuropean cloud; Swiss companyNative Python client; OpenAI compatibility not stated

Two corrections matter for buyers reading older listicles. Featherless’s $25 plan is explicitly limited to human-driven chat and excludes app or API traffic, background automation and benchmarking; its API is billed per token through monthly credits. Giotto’s official API page now lists per-token prices (€2.10 input and €10.50 output per million for Giotto Alto), and no flat CHF plan appears there. An earlier announcement is not a purchasable plan.

Usage-billed platforms compared

These ten platforms do not sell a flat monthly token allowance. They bill per token, per second or per prepaid credit. The “monthly token cap” column is therefore about disclosure, not generosity: a metered platform usually has no monthly cap because every token is charged.

Usage-billed API platforms: billing model, monthly cap, overage, concurrency, context and compatibility
ProviderBilling modelMonthly token capOverageFair-use / concurrencyContextAPI compatibility
OpenRouterPrepaid credits; 5.5% Standard or 8% Business fee on credit purchasesNone publishedNo overage; credits draw downProvider rate limits; 50 requests/day on FreeModel-levelOpenAI-compatible
Together AIPer-token serverless; provisioned throughputNone publishedBilled per token at listAccount and model rate limitsModel-levelOpenAI-compatible
FireworksPer-token serverless; prepaid credits; reserved throughputNone publishedBilled per token at listMonthly spend limits; model rate limitsModel-levelOpenAI-compatible
GroqPer-token Developer tier; free tierNone monthly; daily token caps on FreeBilled per token at month-endRPM, RPD, TPM and TPD limitsModel-levelOpenAI-compatible
CerebrasPer-token Developer tier; prepaid creditsNone publishedBilled per token at listRate limits by model and tierModel-levelOpenAI-compatible
DeepInfraPer-token or per-second; no long-term contractNone publishedBilled per token at listAccount rate limitsModel-level; up to 1,024K on some modelsOpenAI-compatible
MistralPer-token; batch 50% off; cached input discountsNone publishedBilled per token at listTier and model rate limitsModel-levelMistral API; OpenAI migration path
OpenAIPer-token; Standard, Batch, Flex, Fast and Ultrafast modesNone publishedBilled per token at listUsage-tier rate limitsModel-levelNative Responses and Chat Completions
AnthropicPer-token; batch 50% offNone publishedBilled per token at listUsage-tier rate limits; optional monthly spend limitsModel-level; up to 1M varies by modelNative Messages API; OpenAI SDK layer
GoogleFree tier plus per-token paid tierNone publishedBilled per token at listFree and paid tier rate limits; provisioned throughputModel-level; up to 1MGemini API; OpenAI-compatible endpoint

Which providers actually offer no monthly token cap?

If the question is strictly “no monthly token cap”, three groups qualify, and only the first group also combines a flat price with unrestricted API use across more than one model.

  1. True flat API with no monthly token meter: Plugsky Builder and Scale, CheapestInference during an active 8-hour block, and AI2You on its single live model. Each still limits something: Plugsky limits concurrency and RPM, CheapestInference allows one request at a time and declines requests outside the block, and AI2You sells one model per plan.
  2. Flat price with a request quota: UNLI, APIClaw and Awan. Tokens inside a request are not counted, but monthly or daily request counts are. APIClaw also weights heavier models as two or three requests. These are predictable only if your request size stays roughly constant.
  3. Usage-billed with no monthly cap: OpenRouter, Together, Fireworks, Groq, Cerebras, DeepInfra, Mistral, OpenAI, Anthropic and Google. There is no monthly token ceiling because there is no included token allowance; the effective cap is your budget or spend limit.

Featherless and Giotto are often listed as flat-rate providers, but the current official pages do not support that description for production API traffic. Featherless’s flat plan is chat-only; Giotto publishes per-token API prices. A listicle that has not been rechecked can be six months out of date.

Plugsky is one of the few providers that combines (a) a true flat monthly API price, (b) no monthly token cap and no per-token overage on Builder and Scale, (c) 39 models behind one OpenAI-compatible endpoint and (d) live regions in the US, EU, GCC and APAC plus private deployment. That is a scope claim, not a throughput or price claim. On raw price per token for a small model, metered providers are often cheaper.

What “unlimited” must disclose

“Unlimited” is only useful when it is bounded. A provider that will not publish the bounds is asking you to accept an unquantified risk. These are the ten disclosures to require before signing:

  1. What is unlimited: tokens, requests, or both. “Unlimited tokens” on a request-capped plan is a different product from “unlimited requests”.
  2. Monthly token cap: the number, the reset date and whether unused allowance rolls over.
  3. Overage behaviour: per-token charge, hard stop, throttling, or plan upgrade. Plugsky has no per-token overage on Builder or Scale.
  4. Daily caps: a plan can have no monthly cap and still stop at a daily token or request limit.
  5. Concurrency: how many requests can be in flight at once. This is the limit most agent workloads hit first.
  6. RPM and TPM: requests and tokens per minute, including burst behaviour and queueing.
  7. Context and output ceilings: the maximum input and output per request, and whether the plan ceiling overrides the model’s own window.
  8. Model access: all models, a named pool, or a single model. Check the exact IDs and whether legacy IDs map to a different engine.
  9. Fair-use enforcement: what triggers a warning, throttle or suspension, and whether there is an appeal path.
  10. Commercial terms: price-change notice, annual prepay, refund window, SLA exclusions and support tier.

Plugsky’s definition. Unlimited means no monthly token cap and no per-token overage, not infinite capacity. Builder and Scale publish 10 or 16 concurrent requests, 600 or 1,200 RPM, and 256K or 512K input ceilings. Hobby and Starter publish daily caps instead.

Cost scenarios: 10M, 100M and 1B tokens per month

The arithmetic below uses four published per-token rates from Fireworks, Anthropic and OpenAI list prices, an 80/20 input-output token mix, and 2,000 tokens per request. It excludes tax, annual prepay, batch discounts, prompt caching, retries and failed requests. Flat-plan prices are monthly list prices; annual billing is 20% off but is not applied here.

  • Value tier: $0.05 input / $0.20 output per 1M tokens, blended to $0.08 per 1M at 80/20.
  • Mid tier: $0.30 / $1.20, blended to $0.48 per 1M.
  • Frontier-mid tier: $2.00 / $10.00, blended to $3.60 per 1M.
  • Frontier tier: $3.00 / $15.00, blended to $5.40 per 1M.
Monthly cost at 10M, 100M and 1B tokens under four per-token blends versus Plugsky flat plans
Volume / monthRequests at 2,000 tokensValue $0.08/MMid $0.48/MFrontier-mid $3.60/MFrontier $5.40/MBuilder $60Scale $120
10M tokens5,000$0.80$4.80$36$54$60$120
100M tokens50,000$8$48$360$540$60$120
1B tokens500,000$80$480$3,600$5,400$60$120

Break-even is the plan price divided by the blended rate. Builder breaks even at about 11M tokens on the frontier blend, 17M on frontier-mid, 125M on mid and 750M on value. Scale breaks even at about 22M, 33M, 250M and 1.5B respectively. The live claim that Builder breaks even at roughly 250M tokens uses a $0.24 per million blend, which corresponds to an 80/20 mix at $0.15 input and $0.60 output.

The scenario also exposes a capacity issue that token caps hide. One billion tokens a month is about 385 tokens per second sustained. A single-request plan cannot serve that regardless of its token policy. Requests at 2,000 tokens average 16,667 per day, or about 11.6 per minute; concurrency matters more than the monthly total once requests overlap.

Compare like with like. At 1B tokens, Builder costs $60 and the mid-tier per-token bill is $480, but the mid tier is a cheaper model class. On the frontier blend, the per-token bill is $5,400 and Builder is $60. The flat plan is not universally cheaper; it is cheaper when you need expensive models at volume and cannot absorb variance.

When per-token is cheaper

Per-token wins in five common situations. A fair comparison should say so before recommending a flat plan.

  • Below break-even. Under about 11M tokens a month at frontier rates, or 125M at mid rates, Builder costs more than the metered bill.
  • Cheap small models. At $0.08 per million blended, 1B tokens costs $80; Scale at $120 is not the cheaper option.
  • Batch and cache discounts. Batch pricing is commonly 50% off, and cached input can be 90% cheaper on some Mistral models. A flat plan does not pass those discounts through.
  • One specific model or feature. If you need a provider-native audio, fine-tuning, Assistants or Responses feature that a flat catalogue does not serve, staying metered is simpler than splitting traffic.
  • Spiky or seasonal volume. If you use 2M tokens in a quiet month and 200M in a launch month, a metered bill tracks the work; a flat plan charges in the quiet month too.

Flat-rate is the better fit when prompts are long, agents resend context every turn, several models share one key, concurrency above four matters, and the finance team needs an invoice that does not move. The decision is not “flat versus cheap”; it is “predictable capacity versus marginal price”.

What we did not verify

  • We did not run independent latency, throughput or uptime tests on Plugsky or any competitor. No tokens-per-second figure on this page is a benchmark.
  • We did not audit invoices, taxes, FX, annual prepay or enterprise discounts. The scenarios are list-price arithmetic.
  • We did not review a SOC 2 report, penetration test or subprocessor list for any provider named here.
  • We did not test how strictly fair-use policies are enforced at peak, or whether “unlimited” plans throttle under sustained load.
  • We could not load a current Groq price table: groq.com/pricing redirects to the homepage. Groq’s billing model and limits are taken from its official docs, not from a price table.
  • We could not load openai.com/pricing (HTTP 403 for automated requests); OpenAI pricing is taken from platform.openai.com/docs/pricing.
  • Giotto’s 49.90 CHF flat plan is reported by a third-party list but is not published on Giotto’s official API page, which lists per-token prices. We treat the flat plan as unverified and not currently purchasable.
  • Featherless’s flat plan is chat-only by its own terms; we did not verify it for production API use.

Frequently asked questions

Is there an AI API with unlimited tokens and no monthly cap?

Yes, but only on defined terms. Plugsky Builder ($60/mo) and Scale ($120/mo) have no monthly token cap and no per-token overage. Concurrency (10 or 16 requests), requests per minute (600 or 1,200) and per-request context still apply. CheapestInference and AI2You also sell flat API access, with time-block or single-model limits.

Which Plugsky plans have no monthly token cap?

Builder at $60/month and Scale at $120/month. Builder includes 10 concurrent requests, 600 RPM, 256K input and 32K output. Scale includes 16 concurrent requests, 1,200 RPM, a 512K plan input ceiling and 32K output. Hobby ($8) and Starter ($20) have daily token caps of 2M and 5M.

Is unlimited the same as infinite capacity?

No. Plugsky defines unlimited as no monthly token cap and no per-token overage, not infinite capacity. Fair-use concurrency, requests per minute, context ceilings and maximum output still apply. A plan that promises unlimited tokens without publishing those limits cannot be capacity-planned.

Do any other providers offer a true flat-rate API?

Yes, with caveats. CheapestInference sells 8-hour usage blocks with one request at a time. AI2You sells one model per $50 plan. UNLI, APIClaw and Awan combine a flat price with monthly or daily request quotas. Featherless's $25 flat plan is chat-only, and Giotto now publishes per-token API prices.

What is the difference between flat-rate and prepaid credits?

A flat-rate plan fixes the monthly price and does not meter tokens, subject to fair use. Prepaid credits are a balance: each request deducts the provider's per-token price, so the bill is fixed only if you cap the balance. Credits suit spiky or low-volume work; flat plans suit steady production.

How much do 10M, 100M or 1B tokens cost on a flat plan?

On Plugsky, Builder costs $60/month and Scale $120/month at every volume, with no token overage. At an 80/20 input-output mix, a $0.48 per million blended per-token rate costs $4.80 at 10M, $48 at 100M and $480 at 1B. Builder breaks even against that rate at about 125M tokens.

When is per-token billing cheaper than a flat-rate plan?

Per-token is cheaper below the break-even volume, with low-cost small models, with batch discounts, or when you need one provider-specific feature. At an 80/20 mix, Plugsky Builder breaks even at about 11M tokens on a $5.40 per million frontier rate, 17M at $3.60 and 125M at $0.48.

Does a flat-rate plan include every model?

Not always. Plugsky paid plans include all 39 public model IDs. AI2You includes one model per plan. CheapestInference includes the models in your pool. Featherless's flat chat plan covers its catalogue for human chat only. Always check whether the plan covers the model you need at the context you need.

What must a provider disclose before calling a plan unlimited?

It must disclose whether the limit applies to tokens, requests or both; any monthly or daily token cap; overage or hard-stop behaviour; concurrency; RPM or TPM; context and output ceilings; model access; fair-use throttling; and price-change, refund and SLA terms. Without those numbers, unlimited is a slogan, not a specification.

Related pages

Price the workload, not the token

Start on the free plan, then move to Builder or Scale when concurrency and predictability matter more than marginal token price.

Start Free → See pricing

Sources

Every provider claim below was checked on the linked official page on 20 November 2026. Prices and limits change; recheck before quoting.

Official pages used for this comparison
ProviderPage checkedChecked
Plugskyplugsky.com/pricing and /features/unlimited-api20 Nov 2026
CheapestInferencecheapestinference.com/pools20 Nov 2026
AI2Youai2.you20 Nov 2026
UNLIunli.dev20 Nov 2026
APIClawapiclaw.biz/flat-rate-ai-api20 Nov 2026
Featherlessfeatherless.ai/pricing20 Nov 2026
Awan LLMawanllm.com/pricing20 Nov 2026
Giottogiotto.ai/api20 Nov 2026
OpenRouteropenrouter.ai/pricing20 Nov 2026
Together AItogether.ai/pricing20 Nov 2026
Fireworksfireworks.ai/pricing20 Nov 2026
Groqconsole.groq.com/docs/rate-limits and /billing-faqs20 Nov 2026
Cerebrascerebras.ai/pricing20 Nov 2026
DeepInfradeepinfra.com/pricing20 Nov 2026
Mistralmistral.ai/pricing20 Nov 2026
OpenAIplatform.openai.com/docs/pricing20 Nov 2026
Anthropicanthropic.com/pricing20 Nov 2026
Googleai.google.dev/gemini-api/docs/pricing20 Nov 2026
Third-party list (context only)aimultiple.com/flat-rate-llm-api, updated 7 Aug 2026; not used for prices20 Nov 2026

Last checked 20 November 2026. Plugsky’s canonical plan numbers live at /pricing; this page explains the category and does not replace the plan terms.