---
title: "KI-Tarife — was Sie zahlen und wann die API günstiger ist · Undominated.ai"
canonical: https://undominated.ai/de/plans/
description: "Vergleichen Sie 22 KI-Tarife nach echtem Preis und echten Nutzungslimits, inklusive des Token-Volumens, ab dem die Zahlung pro Token die monatliche Pauschale unterbietet."
---

# KI-Tarife — was Sie zahlen und wann die API günstiger ist · Undominated.ai

> Vergleichen Sie 22 KI-Tarife nach echtem Preis und echten Nutzungslimits, inklusive des Token-Volumens, ab dem die Zahlung pro Token die monatliche Pauschale unterbietet.

# Abonnements

22 Tarife für Endkunden und Prosumer, nach Preis und nach dem, was sie tatsächlich unterscheidet: **die Nutzungslimits**, in den Worten des Anbieters. Jede Zeile zeigt außerdem das Token-Volumen, bei dem die Zahlung pro Token dasselbe kosten würde.

 Break-even gemessen gegen Claude Opus 5 — Anthropic Claude Fable 5 — Anthropic Grok 4.6 — xAI GPT-5.6 Sol — OpenAI Kimi K3 — Moonshot AI GLM 5.3 — Z.ai Qwen3.8 Max — Qwen Qwen3.8 2.4T A95B — Qwen Claude Opus 4.8 — Anthropic Muse Spark 1.2 — Meta Nutzungsform Ausgewogen — 3 Token rein pro 1 raus Zusammenfassen — Riesige Eingabe, winzige Ausgabe Chat — Langer, zwischengespeicherter System-Prompt Code — Ausgabe überwiegt Agenten — Kontext wird in jeder Runde neu gesendet

= $11.65/M Token

 - OpenAI ChatGPT Free zum Stack hinzufügen ChatGPT Free OpenAI free kostenlos kein Break-even: es ist kostenlos "Unlimited text chats with GPT-5.6 Luna", asterisked as "Usage must be reasonable and comply with our policies" and "subject to abuse guardrails" — so unlimited means untimed, not unmetered. Everything expensive is explicitly capped: "Limited messages with uploads", "Limited and slower image generation", "Limited voice chats", "Limited deep research", "Limited memory and context", "Limited Codex access". The hard number that actually bites is context: GPT Instant total context window is 27K tokens (~12 pages of input), and the GPT Reasoning window is listed only as "Varies". No access to GPT-5.6 Sol, Sol Pro, or Terra. Unlimited GPT-5.6 Luna text chat
- Web, iOS, Android
- Limited deep research and voice
- 27K GPT Instant context window
 Quelle
- Anthropic Claude Free zum Stack hinzufügen Claude Free Anthropic free kostenlos kein Break-even: es ist kostenlos Metered by a rolling five-hour session window. Anthropic deliberately publishes no fixed message count: the allowance varies with message length, file size, conversation length, model, feature and current demand, so any specific number quoted elsewhere is a third-party estimate. Free does not reach the Opus-class models. Web, desktop and mobile apps
- File and image upload
- Five-hour rolling usage window
 Quelle
- GitHub GitHub Copilot Free zum Stack hinzufügen GitHub Copilot Free GitHub free kostenlos kein Break-even: es ist kostenlos "Limited to 2000 completions per month", a limited AI-credit allowance, and chat with limitations. Crucially, model choice is "Auto selection only" — you cannot pin a model, so you cannot guarantee a frontier model on any given request. Credit counters reset monthly on the 1st at 00:00:00 UTC and do not roll over. 2,000 completions/month
- Limited chat
- Auto model selection only
 Quelle
- GitHub GitHub Copilot Pro zum Stack hinzufügen GitHub Copilot Pro GitHub pro $10 /Monat 0.9M Token/Monat bis Break-even · ~644k Wörter Unlimited code completions plus 1,500 AI credits/month, split 1,000 base + 500 flex. Across every Copilot tier the credit allowance is exactly the plan price x 100, so one AI credit is worth $0.01 and the plan is effectively $10 for $15 of metered AI — the only tier where you get more credit value than you pay. Credits reset on the 1st at 00:00:00 UTC with no rollover; when exhausted you keep working on an included model for the rest of the month. Under the legacy premium-request scheme this tier was 300 premium requests/month with overage at $0.04/request. Opus-class and GPT-5.5/5.6 Sol are NOT available on Pro. Unlimited completions
- 1,500 AI credits/month
- Model selection
- Sonnet-class and GPT-5.6 Terra access
 Quelle
- Mistral Le Chat Pro zum Stack hinzufügen Le Chat Pro Mistral pro $14.99 /Monat 1.3M Token/Monat bis Break-even · ~965k Wörter $14.99/month excluding tax, with "expanded message and search allowances" and "all-day coding access" — no numeric caps published. The distinguishing feature is that the subscription bundles $30/month of Mistral API credits, double the subscription price, which no other consumer plan here does; the free tier bundles $10/month. Verified students pay $5.99. Team is $24.99/user/month with a $50/month minimum spend and 30 GB storage per user. $30/month bundled API credits
- Expanded messages and web search
- All-day coding access
- 100+ connectors
 Quelle
- Microsoft Microsoft 365 Copilot (Business add-on) zum Stack hinzufügen Microsoft 365 Copilot (Business add-on) Microsoft enterprise $18 /Monat $216/Jahr angegeben 1.5M Token/Monat bis Break-even · ~1159k Wörter An add-on to an existing Microsoft 365 subscription, not a standalone product — quoted at $18.00/user/month paid yearly (promotional, down from $21.00), recorded here as $216/user/year; monthly billing runs roughly 20% higher. Microsoft publishes no per-user request or message cap for the add-on. Separately, Microsoft 365 Copilot Chat is free for all Microsoft Entra account users on eligible M365 subscriptions, which makes the paid add-on specifically about Copilot inside Word/Excel/PowerPoint/Outlook/Teams plus agents, not about chat access. Bundled SKUs also exist: Business Premium with Copilot at $32.00/user/month yearly, Business Standard with Copilot at $23.50. Copilot in Word, Excel, PowerPoint, Outlook, Teams
- Reasoning AI for research and data analysis
- Pre-built agents
- Usage analytics
 Quelle
- GitHub GitHub Copilot Business zum Stack hinzufügen GitHub Copilot Business GitHub team $19 /Monat 1.6M Token/Monat bis Break-even · ~1223k Wörter $19 per seat/month with 1,900 AI credits per user per month — the same price x 100 ratio as every other Copilot tier. Buys premium model access plus organisation controls, but note it sits between individual Pro (1,500 credits) and Pro+ (7,000) on allowance, so a developer moving from Pro+ to a Business seat loses credit headroom. Copilot Enterprise is $39/seat with 3,900 credits and priority access to premium models. 1,900 AI credits per user
- Premium model access
- Org policy and admin controls
- Unlimited completions
 Quelle
- Google Google AI Pro zum Stack hinzufügen Google AI Pro Google pro $19.99 /Monat 1.7M Token/Monat bis Break-even · ~1287k Wörter Approximately 4x the free tier's usage limits, with extended access to Gemini 3.1 Pro and Deep Research. Google expresses every AI tier as a multiplier over Free rather than as message counts, so there is no absolute cap to quote. Bundles 5 TB of Google One storage, which is a real part of the value and not an add-on. Note the free tier already includes Gemini 3.6 Flash plus limited 3.1 Pro, so Pro buys volume and Deep Research more than it buys model access. nicht verifiziert ~4x free usage limits
- Gemini 3.1 Pro and Deep Research
- Gemini in Gmail, Docs and Vids
- 5 TB storage
 Quelle
- Poe (Quora) Poe Premium zum Stack hinzufügen Poe Premium Poe (Quora) pro $19.99 /Monat $229.99/Jahr 1.7M Token/Monat bis Break-even · ~1287k Wörter Poe is the one plan here with a fully explicit, published quota: 660,000 compute points per month. The full published ladder is 10,000 points/day (Starter), then 660K, 1.65M, 3.3M and 8.25M points/month, with annual billing saving 17%. Points are a single currency spendable across every model on the platform, and different models consume wildly different rates per message, so the point figure is only meaningful once you know your model mix — a Fable-class message costs many multiples of a small open-weight one. Extra points can be purchased mid-cycle. Subscribing also unlocks each bot's maximum input size and chat history, which is a real limit on the free tier. Caveat: the pricing page rendered euro amounts while stating "Prices shown in US dollars", so the USD figures here are third-party reported and only the point allowances are verified. nicht verifiziert 660,000 compute points/month
- One subscription across many vendors' models
- Maximum input size and chat history unlocked
- Purchasable top-up points
 Quelle
- OpenAI ChatGPT Plus zum Stack hinzufügen ChatGPT Plus OpenAI plus $20 /Monat 1.7M Token/Monat bis Break-even · ~1288k Wörter The tier where reasoning models unlock: "Advanced reasoning models with GPT-5.6". Context roughly doubles over Free on the instant path and jumps ~9x on the reasoning path — GPT Instant total context 54K (~40 pages of input), GPT Reasoning total context 256K (~320 pages). Response times listed as "Fast" rather than Free's "Limited on bandwidth & availability". Deep research, Codex, memory and uploads are all "Expanded" but no numeric cap is published; the page just links to "Limits apply". GPT-5.6 Sol is NOT included (Pro only); GPT-5.6 Terra is unlimited*. Widely-reported but unverified figure: 160 messages per 3 hours on the flagship model. nicht verifiziert GPT-5.6 reasoning models
- 256K reasoning context window
- Projects, scheduled tasks, custom GPTs
- Expanded Codex
 Quelle
- Anthropic Claude Pro zum Stack hinzufügen Claude Pro Anthropic pro $20 /Monat $204/Jahr 1.7M Token/Monat bis Break-even · ~1288k Wörter Two independent meters, and you can hit either: a rolling five-hour session limit, plus a weekly cap that resets at a fixed time assigned to your account. Roughly 5x Free per session. The important structural detail for Pro specifically: Opus-class and Sonnet-class usage draw from a SHARED pool, so one long Opus session eats the rest of your window for every model — on Max they are separate buckets. Includes Claude Code, which consumes the same pool as chat. Annual billing is quoted as $17/month, recorded here as $204/year. ~5x Free usage
- Claude Code included
- Projects
- Five-hour + weekly dual metering
 Quelle
- Perplexity Perplexity Pro zum Stack hinzufügen Perplexity Pro Perplexity pro $20 /Monat $200/Jahr 1.7M Token/Monat bis Break-even · ~1288k Wörter Per Perplexity's own help centre (updated 2026-08-18): "Extended access to Pro Search", a "limited amount [of] Create files and apps queries every 30 days", access to advanced third-party models, image and video generation, increased rate limits on file uploads, and "up to 50 file uploads per project". Perplexity does not publish a numeric Pro Search per-day figure any more — the free tier is described only as "practically unlimited basic searches" with a "very limited amount of Pro Searches". Education Pro is the same plan at $10/month with SheerID verification. The $20/$200 consumer price is NOT stated in the help centre and comes from third-party reporting. nicht verifiziert Extended Pro Search
- Advanced third-party models
- Image and video generation
- 50 file uploads per project
 Quelle
- Cursor Cursor Pro zum Stack hinzufügen Cursor Pro Cursor pro $20 /Monat 1.7M Token/Monat bis Break-even · ~1288k Wörter Cursor meters TWO separate pools that both reset on the monthly billing cycle. The "Cursor Models" pool (Composer, Grok) is described only as "generous included usage" with no published number. The "Other Models" pool — frontier models from Anthropic, OpenAI, Google — is denominated in dollars, and Pro includes $20 of it, exactly the subscription price. Once either pool is exhausted you either pay on-demand "at the same API rates with pay-as-you-go billing" or upgrade. So Pro is best understood as $20 of frontier-model API credit plus unmetered access to Cursor's own models. $20 Other Models pool
- Generous Cursor Models usage
- MCPs, skills and hooks
- Cloud agents
 Quelle
- Anthropic Claude Team (Standard seat) zum Stack hinzufügen Claude Team (Standard seat) Anthropic team $25 /Monat $240/Jahr 2.1M Token/Monat bis Break-even · ~1609k Wörter Per-seat plan billed at $25/seat/month, or $20/seat/month on annual billing (recorded here as $240/seat/year). A Premium seat exists at $125/month or $100/month annual, which is the tier that carries meaningfully higher Claude Code allowances. Enterprise is quoted as seat price plus usage billed at API rates, with a $20/seat base. Central billing and admin
- Shared projects
- Premium seat upgrade path
- Collaboration features
 Quelle
- xAI SuperGrok zum Stack hinzufügen SuperGrok xAI pro $30 /Monat 2.6M Token/Monat bis Break-even · ~1931k Wörter xAI publishes no consumer pricing or limits page that could be reached — x.ai/grok is marketing copy with no prices and grok.com/pricing returns 404, so both the price and the limits here are third-party reported. Reported structure: paid Grok plans moved in mid-2026 to a single shared weekly usage pool spendable across Chat, Imagine, Voice and Build, retiring the older per-two-hour prompt caps that most guides still quote. SuperGrok Heavy is reported at $300/month with the highest rate limits, and X Premium+ at $40/month bundles Grok with X platform perks. Treat every number in this record as unconfirmed. nicht verifiziert Higher Grok limits than free
- Multi-agent reasoning
- Grok Imagine image and video
- Voice mode
 Quelle
- GitHub GitHub Copilot Pro+ zum Stack hinzufügen GitHub Copilot Pro+ GitHub pro $39 /Monat 3.3M Token/Monat bis Break-even · ~2511k Wörter Unlimited completions plus 7,000 AI credits/month (3,900 base + 3,100 flex) — again ~price x 100, so about $70 of metered AI for $39. This is the cheapest tier that unlocks the genuinely expensive models: Claude Opus 4.7/4.8/5, Claude Fable 5, GPT-5.5 and GPT-5.6 Sol are all gated above Pro. Because those models burn credits fastest, the effective request count is far lower than the 7,000 figure suggests. Legacy scheme: 1,500 premium requests/month. 7,000 AI credits/month
- Opus 4.8/5 and Fable 5 access
- GPT-5.6 Sol access
- Copilot coding agent
 Quelle
- Google Google AI Ultra (5x) zum Stack hinzufügen Google AI Ultra (5x) Google max $99.99 /Monat 8.6M Token/Monat bis Break-even · ~6437k Wörter 5x the AI Pro usage limits, 10,000 Google Flow credits per month for video/creative generation, and storage starting at 20 TB. Google cut Ultra's entry price from $250 to about $100/month at I/O 2026 and split the old single tier into 5x and 20x variants; the 20x sits near $199.99 with 25,000 Flow credits. Flow credits are the binding constraint for video work, not the chat multiplier. nicht verifiziert 5x Pro limits
- 10,000 Flow credits/month
- 20 TB+ storage
- YouTube Premium
 Quelle
- OpenAI ChatGPT Pro zum Stack hinzufügen ChatGPT Pro OpenAI pro $100 /Monat 8.6M Token/Monat bis Break-even · ~6438k Wörter Sold as "From" a base price with "5x or 20x more usage" — i.e. two Pro sub-tiers at the same feature set, differing only in volume, which is why priceMonthlyUsd records the entry point. Unlocks "Pro reasoning with GPT-5.6 Sol Pro" and unlimited* GPT-5.6 Sol, neither of which Plus gets at all. Largest context of any consumer tier: GPT Instant 128K (~250 pages of input), GPT Reasoning 400K (~680 pages). "Maximum Codex tasks", "Maximum deep research", "Unlimited and faster image creation", all asterisked "Unlimited subject to abuse guardrails". nicht verifiziert GPT-5.6 Sol Pro reasoning
- 400K reasoning context window
- 5x or 20x usage sub-tiers
- Maximum deep research and Codex
 Quelle
- Anthropic Claude Max (5x) zum Stack hinzufügen Claude Max (5x) Anthropic max $100 /Monat 8.6M Token/Monat bis Break-even · ~6438k Wörter Quoted on the pricing page as "From $100/month". Delivers five times Pro's usage per five-hour session; the weekly cap and other capacity controls still apply on top, so 5x per session is not 5x per week. Unlike Pro, Opus-class models get their own weekly reset separate from the all-other-models reset, so heavy Opus use no longer cannibalises Sonnet headroom. Anthropic publishes no hours-per-week figure — the settings page shows your real consumption. 5x Pro usage per session
- Separate Opus and non-Opus weekly buckets
- Higher Claude Code limits
- Priority access at peak times
 Quelle
- Anthropic Claude Max (20x) zum Stack hinzufügen Claude Max (20x) Anthropic max $200 /Monat 17.2M Token/Monat bis Break-even · ~12876k Wörter Twenty times Pro usage per five-hour session, same dual-meter structure as Max 5x with separate Opus and non-Opus weekly resets. The price is NOT stated on claude.com/pricing, which only says Max starts "From $100"; the $200 figure comes from third-party reporting and is the reason this record is marked inferred while Max 5x is verified. nicht verifiziert 20x Pro usage per session
- Separate Opus and non-Opus weekly buckets
- Highest Claude Code allowance
 Quelle
- Perplexity Perplexity Max zum Stack hinzufügen Perplexity Max Perplexity max $200 /Monat $2000/Jahr 17.2M Token/Monat bis Break-even · ~12876k Wörter Everything in Pro plus, in Perplexity's wording, "the highest level access to our full suite of advanced AI models" with "minimal limits on volume or frequency" for Create files and apps, faster access to new OpenAI and Anthropic releases in Perplexity Research, early access to new products, priority support with dedicated infrastructure, and Brain — a self-improving memory system for Perplexity Computer, free during the current research preview. Still no published numeric caps. Price is third-party reported, not in the help centre. nicht verifiziert Minimal limits on Create files and apps
- Highest model access tier
- Brain memory system (preview)
- Early access
 Quelle
- Cursor Cursor Ultra zum Stack hinzufügen Cursor Ultra Cursor max $200 /Monat 17.2M Token/Monat bis Break-even · ~12876k Wörter $200/month for a $400 Other Models pool — the only plan in this dataset where the included frontier-model allowance is worth double the subscription price, versus 1x at Pro ($20/$20) and ~1.17x at Pro+ ($60/$70). Plus the same unmetered "generous" Cursor Models pool. Both pools reset monthly and overage bills at API rates. That 2x ratio, not a request count, is the actual reason to buy Ultra. $400 Other Models pool for $200
- Generous Cursor Models usage
- Priority feature access
- Cloud agents
 Quelle

Der Break-even nimmt das gewählte Modell in der gewählten Nutzungsform an. Es ist der Punkt, an dem die Abrechnung pro Token dasselbe kostet wie das Abo: darunter zahlen Sie für Bequemlichkeit, darüber ist das Abo die günstigere Option. Anbieter veröffentlichen diese Zahl nicht.