Check your actual bill →

For power users · honest comparison · 2026-05-22

Your AI subscription is renting you the wrong thing.

You pay $20 to $200 a month and somewhere between the third rate-limit popup, the KYC re-verification email, and the "switch to GPT-5.5 to try a different angle" you can't do, it stops looking like a tool and starts looking like a leash. The frontier subscriptions are great products. They are also locked, capped, and tied to your real name. This page is the comparison the vendors would rather you didn't read.

200K tokens · no card · no KYC · cancel by not renewing.

The friction wall: six things your subscription does to you

Each of these is real, recent, and reproducible. Sources are linked where the vendor publishes a number; the rest come from the operator's 14 months on Claude Pro plus the discord channels of every team that hit the wall before you did.

01 · Rate limit hell

Limits that hit you when you are actually working

Every frontier subscription throttles in 5-hour rolling windows. The throttle is not a soft warning, it is a hard pause in the middle of a refactor.

Claude Max 20x: ~220K tokens per 5-hour window, weekly Opus throttle.
Cursor Pro: usage-based fast-request credits, unpublished monthly count, then "slow mode".
ChatGPT Plus: GPT-5.5 caps that bite around noon Pacific.
02 · KYC and card on file

Every frontier vendor has your real identity on file

Card on file, phone number, IP. The billing trail is a discovery surface. If your work touches a regulator, a competitor, or an NDA-covered customer, that surface is now part of your threat model.

Anthropic, OpenAI, Cursor, Copilot, Codeium: KYC + card required at signup. No Bitcoin path on any of them.
03 · Model lock-in

You cannot escalate Claude to GPT in the same flow

You hit Sonnet's reasoning wall on a thorny diff and you want to ask Opus or GPT-5.5 the same question. With a Claude or OpenAI subscription, you can't. You'd need a second account, a second card, a second login.

Claude Pro: Anthropic models only.
ChatGPT Plus: OpenAI models only.
Cursor Pro: Cursor's curated set, not arbitrary providers.
04 · No BYO keys

You cannot consolidate the spend you already have

You already pay Anthropic direct, OpenRouter, maybe a Together credit. None of the consumer subscriptions let you point them at your existing keys, so you pay twice: once for the subscription, again for the spillover.

Cursor, Copilot, ChatGPT, Claude: no BYO key endpoint. Your usage on them is their usage, on their meter, at their markup.
05 · Quota surprises

You discover the limit by hitting it mid-task

Cursor's "you used your fast quota" notification arrives somewhere between commit and pull request. ChatGPT's "switch to GPT-4o" downgrade message arrives during the demo.

No usage telemetry until you are over. No per-key budget. No per-call cost in the response headers. You learn what a request cost only when the throttle bites.
06 · EU residency that isn't

You cannot actually buy EEA-only inference from them

Anthropic and OpenAI both route through US infrastructure. The DPAs they offer are governed by US law and US discovery. If your customer's procurement form asks for EEA-only processing, the consumer subscriptions cannot honour it.

Anthropic Console: US-resident. OpenAI Platform: US-resident. EU enterprise tiers exist on paper for very large contracts only.

Side by side: the $10 to $20 entry tier

Numbers verified 2026-05-22 against each vendor's published pricing page. Where a vendor publishes a token allowance, we used it; where they publish a request count instead, we noted that. This is the table for the casual-to-prosumer crowd. If you are paying $100+ a month, scroll past this one to the heavy-use matrix below.

ProductPriceToken budgetModelsKYC?Crypto?BYO keys?Cancel anytime?EU residency?Real pain point
Claude Pro$20 / mo~5x free; 5-hour capsAnthropic onlyYesNoNoYesNoLocked to Anthropic; weekly Opus throttle; US-only processing
Claude MaxFrom $100 / mo5x or 20x Pro, your choice at signup; still session-cappedAnthropic onlyYesNoNoYesNoEven the 20x tier still hits 5-hour caps; Agent SDK credit helps but you cannot route around Anthropic
Cursor Pro$20 / moUsage-based fast-request credits; exact monthly count not published, then throttled to slow modeCursor's curated setYesNoPartialYesNoFast-request allowance is not published in raw numbers, making it hard to budget; overage can run well past the base $20; BYO key only for OpenAI/Anthropic
GitHub Copilot Indiv$10 / mo (also Free at $0, Pro+ $39, Max $100)Unlimited completions; chat/premium requests capped and scale with tierGitHub's setYesNoNoYesNoIDE-locked; chat model swaps without notice; four tiers now exist and the jump from $10 to $39 or $100 is steep
ChatGPT Plus$20 / moGPT-5.5 capped, falls back to GPT-4oOpenAI onlyYesNoNoYesNoSilent downgrade to weaker model when caps hit; OpenAI lock-in; no API parity
Windsurf (Codeium) Pro$20 / mo (raised from $15 on Mar-19-2026)Daily + weekly quotas (numeric figures not public)Claude Sonnet 4.6, GPT-5.4, SWE-1.5YesNoNoYesNoQuota system replaced credits Mar 2026; numeric caps undisclosed; meters can blank a workday with no warning
llmdeal Pro$119 / mo50M tokens included, overage $4 / 1MSmart-routed across 6 open-weight models + Frontier add-onNoBTC / XMR / LTCYes, $19 add-on (free on $100+ plans)Yes, by not renewingYes, EEA GPU routing on requestNone of the above. OpenAI-compatible endpoint, your code does not change.

Sources: anthropic.com, openai.com, cursor.sh, github.com/features/copilot, codeium.com, llmdeal.me/pricing.html. Last verified 2026-05-22.

$100+ tier showdown: when Pro subscriptions don't cut it

If your AI bill clears $100 a month, the comparison changes. You are now in territory where Claude Max, ChatGPT Pro, Cursor Business with multiple seats, API-direct spend, and stacked subscriptions all compete. The interesting question is no longer "which subscription is best" but "how many subscriptions am I paying for and what do they actually deliver." Same data discipline as the entry table: every figure traced to the vendor's pricing page or, where blocked by anti-bot, to two aggregator cross-checks. Verified 2026-05-22.

ProductPriceToken budgetModelsRate limitsKYC?Crypto?BYO keys?Cancel?EU residency?Real pain point
Claude MaxFrom $100 / mo (5x); $200 / mo for 20x~88K tokens / 5-hour window on 5x; ~220K tokens / 5-hour window on 20xAnthropic only (Sonnet, Opus, Haiku)5-hour rolling buckets on both tiers; weekly Opus cap; chat and Claude Code share the same bucketCard on fileNoNoYesNoPeak-hour throttling Anthropic openly admits to. Even the 20x tier stays capacity-gated; chat eats the same bucket as Claude Code.
ChatGPT ProFrom $100 / mo5x or 20x Plus's 5-hour bucket, your choice; 20x adds 250 Deep Research runs / moOpenAI only; ~1M token context on 20x (~680 pages)5x or 20x Plus across all model classes, picked at signup; Deep Research a separate monthly cap on 20xCard on fileNoNoYesNoThe "5x/20x Plus" framing is relative, not absolute — you don't know the underlying token volume. Still capacity-gated at peak on either tier. No refund outside EU/UK/Turkey. No crypto.
Perplexity Max$200 / mo ($167 / mo if paid annually)Unlimited Labs + Perplexity Computer; 300+ Pro Searches/day baselineMulti: GPT-5.5, Sonnet 4.6, Sonar Large, Grok"Unlimited" Labs (soft-capped); daily Pro Search ceiling unchanged from ProCard on fileNoNoYesNoSearch-first product. Useful for research, weak for sustained code generation. Same $200 ceiling as Claude/ChatGPT top tiers.
Google AI UltraFrom $99.99 / mo (5x); $199.99 / mo for 20xGemini 3.1 Pro + Veo + NotebookLM Pro; daily caps not publicGoogle only (Gemini 3.1 Pro)Daily/monthly caps on Veo + Deep Research not enumerated in numeric terms; usage scales with the 5x/20x tier pickedGoogle accountNoNoYesWorkspace data onlyLocked to Google ecosystem, now split into two tiers like Claude Max and ChatGPT Pro. The Workspace bundle inflates perceived value if you don't need 2 TB Drive.
SuperGrok Heavy$300 / moGrok 4 Heavy multi-agent + 256K context + parallel test-time computexAI only (Grok 4 Heavy)Highest priority access; soft caps on image gen ("unlimited" with throttling after 50 to 100 rapid gens)X / grok.com accountNoNoYesNo10x price jump from SuperGrok ($30) to Heavy ($300) with no middle tier. Non-refundable per xAI ToS.
Cursor Teams (3 seats)$40 / user, 3 seats = $120 / moCredit-based per seat (post Jun 2025 overhaul)Cursor's curated set + BYO partial1 req/min, 30/hour API-layer; effective request count cut ~55% in Jun 2025 overhaulCard per orgNoPartialYesNoHN-documented "$350 in one week" overage cases. CEO publicly acknowledged "mishandling" the Jun 2025 rollout. Credit-to-request ratio is opaque.
Copilot Business (5 seats)$19 / user, 5 seats = $95 / mo$19 in AI Credits per seat / mo (post Jun 1 2026 overhaul)Multi: GPT-5.5, Sonnet 4.6, Gemini 3.1, othersCode completion unmetered; chat / agent / premium models burn AI Credits at per-model rateGitHub orgNoNoYesNoPro signups paused April 20 2026. Usage-based billing starts June 1 2026. Users are bracing for overage surprises.
Copilot Enterprise (5 seats)$39 / user, 5 seats = $195 / mo$39 in AI Credits per seat / moMulti-model + custom org modelsPer-seat AI Credit budget; org-level overageGitHub orgNoNoYesNoDoubles the per-seat cost vs Business for SSO, audit logs, and policy controls. Pricing model changes June 1 2026.
Codeium Teams (4 seats)$80 team fee + $40 / seat, 4 seats = $240 / moDaily + weekly quotas per seat (numeric figures undisclosed)Sonnet 4.6, GPT-5.4, SWE-1.5Quota meters refresh automatically; no numeric figures published since March 19 2026 overhaulCard per orgNoNoYesNoQuota meters can blank a workday with no warning. Numeric caps are not disclosed.
Tabnine Enterprise (5 seats)$39 / user, 5 seats = $195 / moNot strictly metered; bandwidth-limitedMulti (Claude, GPT, Mistral) on Enterprise; in-house on ProSoft bandwidth limitsBusiness verification on enterprise contractsNoAir-gapped option on EnterpriseAnnual on EnterprisePrivate deployment optionMulti-model only at Enterprise tier. In-house Tabnine models perceived as weaker than frontier in Reddit threads.
OpenAI API direct (100M tokens / mo)~$3,000 / mo (50M in @ $10 + 50M out @ $50, flagship rate)Pay-per-token, no monthly bundleGPT-6 Astra ($10/M in, $50/M out, flagship); GPT-5.6 Sol promo tier ($4/M in, $20/M out)Tier 1-5 system; new accounts hit aggressive RPM/TPM caps; spend history unlocks higher tiersAccount verify for Tier 4/5NoDirect APIN/ANoGPT-6 Astra runs $10/$50 per 1M at the flagship tier. Pre-paid credits non-refundable. No crypto.
Anthropic API direct (100M tokens / mo)~$600 / mo (50M in @ $2 + 50M out @ $10) on Sonnet 5 standard ratePay-per-token; batch 50% off; prompt cache up to 90% offSonnet 5 ($2/$10), Opus 5 ($5/$25), Claude Fable 5.1 / Mythos 5.1 ($10/$50, top tier), Haiku 4.5 ($1/$5)Tier system gates higher limits behind cumulative spendAccount verify for higher tiersNoDirect APIN/ANoFable/Mythos 5.1 is 5x the Sonnet 5 price. No crypto. No native consolidation with your OpenAI / Together / OpenRouter spend.
OpenRouter (100M tokens / mo)~$950 to $1,320 / mo (Sonnet 4.6 + 5.5% credit fee)Pay-as-you-go credits300+ models across providersProvider-specific; pooled rate limits across OR routesNoneBTC/ETH via Stripe Crypto (5% fee)Yes (5% surcharge over 1M reqs/mo)N/ANo5.5% credit-purchase fee compounds. Crypto path is Stripe Crypto, not native (technically KYC at the gateway).
Entry stack (Claude Pro + Cursor + Copilot)$50 / moThree separate buckets, no consolidationAnthropic + Cursor's curated set + GitHub's setThree independent throttles to trackThree cards on fileNoNoYes (three cancels)NoThree logins, three bills, three rate limits. Workload doesn't move across them.
Light pro stack (Max 5x + Cursor + Copilot Business)~$159 / mo (Max 5x $100 + Cursor 1 seat $40 + Copilot Business 1 seat $19)Three buckets again, none consolidatedAnthropic + Cursor's set + GitHub's setThree independent throttles, the Max 5x one is the binding constraint for chat workThree cards on fileNoNoYes (three cancels)NoYou are paying for Claude twice over (Max 5x for chat, the Anthropic Claude calls inside Cursor and Copilot). No single point of cost control.
Team stack (Max 20x + Cursor Teams 3 + Codeium Teams 3)~$410 / mo (Max 20x $200 + Cursor Teams 3 seats $120 + Codeium Teams 3 seats $90)Three buckets, three vendor portalsAnthropic + Cursor's set + Codeium's setAll three throttle independentlyThree cards on fileNoNoYes (three cancels)NoThree procurement reviews, three SSO setups, three bills. Workload doesn't move across the boundaries.
llmdeal Pro$119 / mo50M smart-routed tokens; overage $4 / 1M6 open-weight via Groq/Cerebras/NIM + Frontier Credits add-on for Claude/GPT-5.5Fixed-window slot, no peak-hour silent squeezeNoneBTC / XMR / LTCFree on this tierBy not renewingEEA GPU on requestNone of the above. OpenAI-compatible endpoint, your code does not change.
llmdeal Elite$159 / mo100M tokens + 128K context standardAll 6 open-weight + Frontier Credits add-onFixed slot, no degradation under load; reserved capacity headroomNoneBTC / XMR / LTCFree on this tierBy not renewingEEA GPU on requestApples-to-apples vs Claude Max 20x / ChatGPT Pro Max at the same $200 ceiling. Twice the budget and you can swap models mid-request.
llmdeal EU-Sovereign$199 / mo60M EU-only tokens + signed GDPR Article 28 DPAAll 6 open-weight, EEA-resident routing onlyFixed slotNoneBTC / XMR / LTCFree on this tierBy not renewingEEA-only with DPAFor the procurement form that asks "EEA-only processing, signed DPA, no US discovery surface." None of the consumer subscriptions can sign this.
llmdeal Business$499 / mo300M tokens + reserved capacity poolAll 6 open-weight + Frontier Credits add-onReserved slice, no shared-tenancy throttleNoneBTC / XMR / LTCFree on this tierBy not renewingEEA or US, your callFor the 5-to-10 person team currently stacking Cursor Teams, Copilot Business, and Claude Max 5x for a heavy user.
llmdeal Scale$1,999 / mo1.5B tokens + 99.5% SLAAll 6 open-weight + Frontier Credits add-onReserved capacity, formal SLANoneBTC / XMR / LTCFree on this tierBy not renewingEEA or US, your callFor the customer currently spending $900 to $1,750 a month on OpenAI / Anthropic API direct with no slot, no consolidation, no SLA.
llmdeal Consortium Pro$99 / moCommunity-routed inference · 25-50% off open-weight ratesconsortium-llama-70b / consortium-qwen-coder / consortium-qwen-235b — community-hosted boxes, automatic fall-back to LiteLLM if unhealthyrpm=1200 · tpm=5M · max_parallel=200 (highest we ship; consortium pool absorbs the load)NoneBTC / XMR / LTCFree on this tierBy not renewingEEA or US, your callFor developers who want the cheapest possible open-weight inference and don't mind that someone else's GPU under their desk is doing the work. Capacity-limited — gated tier on purpose.
llmdeal Fleet$519 / moAll consortium open-weights at priority-queue throughputSmart-route + Llama 70B / Qwen3-235B / Qwen3-Coder-480B / Nemotron / Maverickmax_parallel=20 · rpm=120 · tpm=200k · priority queueNoneBTC / XMR / LTCFree on this tierBy not renewingEEA or US, your callFor the small team running a real product on open weights who want consolidated billing, priority queue, and 5 separate project keys for staging / prod / mobile / web.
llmdeal Fleet Plus$959 / moFleet + 60M frontier credits/mo (Sonnet / GPT-5.5 / Gemini Pro)Everything in Fleet + the frontier reference setmax_parallel=40 · rpm=240 · tpm=400k · cache-priority · 24h priority supportNoneBTC / XMR / LTCFree on this tierBy not renewingEEA or US, your callProduction SaaS doing agentic coding or RAG where 5–10% of calls genuinely need the frontier and 90% can ride open-weight — Fleet Plus prices both correctly.
llmdeal Burst Day$129 one-time8M tokens, any model — frontier included — within 24h windowFull catalog including Claude Sonnet 4.6 / GPT-5.5 / Gemini Pro24h at Fleet-tier limits · stacks with any subscriptionNoneBTC / XMR / LTCFree on this tierNo-op — one-offEEA or US, your callA Sunday-afternoon batch sweep · a benchmark across all your prompts · a one-day codebase migration. Cheaper than a $99 Claude Pro month for someone who just needs the capacity once.

Sources: claude.com/pricing, chatgpt.com/pricing, perplexity.ai/pro, gemini.google/subscriptions, grok.com/plans, cursor.com/pricing, github.com/features/copilot/plans, windsurf.com/pricing, tabnine.com/pricing, openai.com/api/pricing, platform.claude.com/docs/en/about-claude/pricing, openrouter.ai/pricing, llmdeal.me/pricing.html. Last verified 2026-05-22. Wholesale-token totals computed at the vendor's published per-million rate with a 50/50 input/output split unless otherwise noted; your real workload will skew.

Stacking math: what consolidation actually saves

Three real combinations we see in inbound DMs. Numbers are computed against the same published vendor rates as the matrix above; we round honestly. We are not trying to make every stack look like a save. The third example is a case where llmdeal costs more in dollars and the trade is volume plus consolidation, not raw price.

Stack 1 · solo power user

"I pay Claude Max 5x + Cursor Business 1 seat + Copilot Business 1 seat and still hit caps."

$100 + $40 + $19 = $159 / mo today

Switch to Pro $119 + Frontier Credits add-on $89 = $239 / mo. BYO Keys included free at this tier, so your existing Anthropic key handles the closed-weight prompts directly without a markup hop. Three subscriptions collapse into one OpenAI-compatible endpoint.

+$80 / mo for 50M slot tokens, 5M Frontier credits, and zero peak-hour throttling. Net: same dollar order of magnitude, no rate-limit interruptions, one bill, one key, one set of model swaps.
Stack 2 · 5-person team

"We're a 5-person team paying Cursor Teams + Copilot Business + Claude Max 5x for one heavy user."

$200 (Cursor Teams 5) + $95 (Copilot Business 5) + $100 (Claude Max 5x) = $395 / mo today

Switch to Business $499 / mo. 300M tokens shared across the team, BYO Anthropic key keeps the existing Claude relationship in your name, reserved capacity pool means no peak-hour squeeze. Unified billing, one procurement review.

+$104 / mo for 5x to 10x the model menu, no per-seat math, BYO-included consolidation, and an EEA-resident DPA option none of the entry-tier subscriptions can sign.
Stack 3 · API-direct heavy user

"I'm running 100M Sonnet 4.6 tokens a month directly on the Anthropic API."

50M input @ $3/M + 50M output @ $15/M = ~$900 / mo wholesale, plus tier-system rate limits

Switch to Scale $1,999 / mo. 1.5B tokens (15x the volume), reserved capacity slot, 99.5% SLA, BYO Anthropic key still available for the Sonnet-specific prompts via Frontier Credits add-on, crypto checkout if you prefer. EEA or US-resident routing, your call.

Costs more in raw dollars ($1,999 vs $900) and the trade is 15x volume on open-weight smart-routed inference plus a slot that survives Anthropic's peak-hour squeeze. If your workload is Sonnet-pure and capped at 100M tokens, the API-direct path stays cheaper; if you are looking for headroom and consolidation, this is the swap.

Where llmdeal honestly loses

No comparison page deserves trust until it admits the cases where it isn't the right answer. Here are four.

  • The $20 casual user. If you write a few prompts a day and you want a chat UI on a phone, ChatGPT Plus and Claude Pro at $20 are better than any llmdeal tier under $69. Our entry point is built for people who hit subscription rate limits, not for people who don't.
  • The Artifacts / Canvas user. Claude's Artifacts and ChatGPT's Canvas are genuinely useful as non-API consumption surfaces. llmdeal is an OpenAI-compatible endpoint. You bring your own UI. If a polished consumer-grade chat experience is the product you actually wanted, the consumer subscription is the right buy.
  • IDE-native polish. Cursor's editor integration is tighter than what you get from pointing your own client at our endpoint. We are an API, not an IDE. If you live inside Cursor and the rate cap doesn't bite you, the $20 is well spent there.
  • Pure single-model API workloads. If your workload is "100M Sonnet 4.6 tokens, nothing else" and you are not interested in open-weight routing or consolidation, the Anthropic API at $900 wholesale is cheaper than llmdeal Scale at $1,999. We win on volume + consolidation + slot predictability + crypto + EEA. We do not win on raw $/token against a vendor selling exactly what you wanted at wholesale.

The catalog, in plain English

27 SKUs from $0 to $5,999/mo. Same OpenAI-compatible endpoint across all of them. Pick the bundle that matches your workload, switch tiers by emailing us, cancel by not renewing.

Free Trial - $0 · 200K tokensHobbyist - $9Solo - $29Mix Pack - $49Vision - $59Fast - $69Starter - $75Coder - $99Stack - $100Reasoner - $129Pro - $119 · most popularElite - $159EU-Sovereign - $199US-Sovereign - $199Business - $499Fleet - $519Fleet Plus - $959Scale - $1,999Burst Day - $129 one-time+ Frontier Credits - $89 / mo add-on+ BYO Keys - $19 / mo add-on (free on $100+)

What you get on every tier: no KYC, no card on file, crypto checkout in BTC, XMR, or LTC, OpenAI-compatible endpoint that drops into any OpenAI SDK without code changes, smart routing across Groq, Cerebras, and NVIDIA NIM, your choice of EU-resident or US-resident processing, and the ability to bring your own provider keys to consolidate spend. Open-weight stack means Llama 3.3 70B, Qwen3-235B, GPT-OSS-120B, Nemotron-Super-120B, Llama-4-Maverick, Qwen3-Coder-480B, and more, all swappable in one request. The Frontier Credits add-on (live 2026-06-15) routes specific prompts to Claude and GPT-5.5 when you need the closed-weight frontier on top.

Four customer paths

Worked-out switches for the four combinations we see most often in inbound DMs. Numbers are conservative.

Path 1 · solo dev, entry tier

Cursor Pro $20 / mo and bumping into the post-Jun-2025 request cap

switch to →

Mix Pack $49 with 25M tokens + smart-route across Qwen3-Coder, Llama 3.3 70B, and DeepSeek V3.2, plus the BYO Keys add-on $19 so your existing Anthropic key handles the hard prompts.

$68 / mo total. Same OpenAI-compatible endpoint. Cursor stays in your editor, llmdeal handles the requests. No more silent quota cuts.
Path 2 · solo power user, $100+ tier

Claude Max 5x $100 + Cursor Business 1 seat $40 + Copilot Business 1 seat $19 = $159 / mo

switch to →

Pro $119 + Frontier Credits $89 add-on. 50M smart-routed tokens, 5M Frontier credits for Claude/GPT-5.5, BYO Keys included free.

$239 / mo. +$80 vs the stack, with no peak-hour throttling, no chat-eats-from-Code bucket, one endpoint, one bill, one rate limit. Consolidation, not discount.
Path 3 · 5-person team stacking

Cursor Teams 5 seats $200 + Copilot Business 5 seats $95 + Claude Max 5x for heavy user $100 = $395 / mo

switch to →

Business $499 / mo. 300M tokens shared across the team, reserved capacity pool, BYO Anthropic key included free so the team's existing Claude relationship stays in your org's name.

$499 / mo, +$104 over the stack. The trade: 5x to 10x the model menu, one procurement review, one SSO setup, one bill, an EEA-resident DPA option, and no per-seat math.
Path 4 · API-direct heavy / enterprise

OpenAI API + Anthropic API + Pinecone + Cohere reranker

switch to →

Scale $1,999 with smart-route across the open-weight stack + dedicated rate-limit pool + 99.5% SLA. Below that, Fleet Plus $959 covers production SaaS that needs ~60M frontier tokens/month with cache-priority and 24h-priority support. For one-off heavy days, Burst Day $129 stacks with anything.

Predictable monthly spend, no card on file, BYO support across providers, and you keep the OpenAI SDK you already wrote your code against.

How we actually make money

The trust angle, because every "no KYC crypto" pitch deserves a hard look at the business model.

We do not sell your data. We do not log prompts. We do not run a training pipeline on what you send. We make money the boring way: on the spread between what an open-weight model costs us to host (or to route through a wholesale provider like Groq, Cerebras, or NVIDIA NIM) and what we charge you per token. The Pro tier's 50M-token bundle has roughly a 45 to 55 percent gross margin at current prices, which is below what Anthropic or OpenAI run at, because we are not paying for a brand and we are not paying for ChatGPT.com's consumer surface.

  • No prompt logging. Requests are billed by token count, not stored by content. The dashboard shows you a request count, not a transcript.
  • Open-weight stack. The models we route to are public. You can verify our benchmark claims on ATLAS, LMSYS, and Artificial Analysis. None of those surfaces are operated by us.
  • Crypto-only billing. We do not accept fiat for a reason. No card processor, no chargeback risk, no payment-network KYC requirement we have to honour upstream.
  • BYO Keys is the real product. Pay us for the routing layer, bring your own provider keys, keep the relationship with the upstream vendor in your own name. We are infrastructure, not a reseller of your account.
  • EU or US, your call. Toggle EEA-resident routing in the dashboard and your prompts stay in the EU. We can sign a GDPR Article 28 DPA on Elite and above.
  • Cancel by not renewing. There is no contract. There is a monthly subscription you renew or you don't. Credits never expire.

Try it on $0 first.

200,000 tokens, no card, no KYC, no follow-up email asking for ID. Point any OpenAI SDK at https://llmdeal.me/v1 and you are in.