DeepSeek R1 — Spezifikationen
| Entwickler | DeepSeek |
|---|---|
| Typ | LLM (MoE, Reasoning) |
| Modality | Text → Text |
| Parameter | 671 Mrd. insgesamt / 37 Mrd. aktiv (MoE) |
| Kontextfenster | 128 K |
| Maximale Ausgabe | — |
| Lizenz | MIT (offen) |
| Offene Gewichte | Ja |
| Veröffentlicht | 2025 |
| Eingabepreis | 0,50 $ pro 1 Mio. |
| Ausgabepreis | 2,15 $ pro 1 Mio. |
| API-Anbieter | DeepSeek, DeepInfra, OpenRouter |
Lokal ausführen
| VRAM (4-Bit) | ~400 GB |
|---|---|
| Mindest-GPU | Multi-GPU-Server |
What is DeepSeek R1?
DeepSeek R1 is the landmark open reasoning model — a 671B mixture-of-experts activating
37B parameters per token, MIT-licensed, with a 128K context window. It delivered
frontier-class chain-of-thought reasoning at a fraction of proprietary cost and is the model
that made open reasoning weights a mainstream option rather than a research curiosity.
Two things follow from the architecture. First, the API is genuinely cheap at $0.50 in /
$2.15 out per million tokens, because only 37B of the 671B parameters are active on any given
token. Second, self-hosting is not cheap at all: the full weights need roughly 400 GB of VRAM
even at 4-bit quantisation, which means a multi-GPU server, not a workstation. The MIT
licence gives you the right to run it, but the hardware bill is what decides whether you
will. For most teams the correct reading is that R1 is an API model that happens to be
auditable and portable — you can move providers, inspect the weights, and never be locked in,
without personally operating an eight-GPU node. If you want R1-style reasoning on hardware
you own, the 70B distill is the model to look at instead.
DeepSeek R1 pricing: API cost per 1M tokens
| Eingabe (pro 1 Mio. Token) | $0.500 |
|---|---|
| Ausgabe (pro 1 Mio. Token) | $2.15 |
| Output/input ratio | 4.3× |
| Blended (4:1 in:out) | $0.830 per 1M tokens |
What DeepSeek R1 costs per month
Real monthly spend at a 4:1 input-to-output mix — the ratio a typical chat or RAG workload actually produces.
| Workload | Tokens/Monat | Cost / month |
|---|---|---|
| Nebenprojekt | 1 Mio. Eingabe / 0,25 Mio. Ausgabe | $1.04 |
| Kleines Team | 20 Mio. Eingabe / 5 Mio. Ausgabe | $21 |
| Produktion | 200 Mio. Eingabe / 50 Mio. Ausgabe | $208 |
Run your own numbers in the KI-API-Kostenrechner.
Cheaper alternatives to DeepSeek R1
| Modell | Gewichteter Preis pro Million US-Dollar | You save |
|---|---|---|
| DeepSeek V4-Pro offenere | $0.522 | 37% cheaper |
| DeepSeek V4-Flash offenere | $0.168 | 80% cheaper |
| Llama 4 Maverick offenere | $0.240 | 71 % günstiger |
Self-host or pay the API?
DeepSeek R1 is open-weight, so you can run it yourself. It needs ~400 GB of VRAM at 4-bit (Multi-GPU server). Self-hosting only beats the API once your volume is high enough to keep that hardware busy — the Selbsthosting-vs.-API-Rechner works out the break-even point for your token volume.
Häufig gestellte Fragen
How much does DeepSeek R1 cost per 1M tokens?
DeepSeek R1 costs $0.500 per 1M input tokens and $2.15 per 1M output tokens. At a typical 4:1 input-to-output mix that blends to about $0.830 per 1M tokens.
How much does DeepSeek R1 cost per month?
A small-team workload of 20M input and 5M output tokens a month costs about $21 on DeepSeek R1. A side project (1M in / 0.25M out) costs roughly $1.04.
What is a cheaper alternative to DeepSeek R1?
DeepSeek V4-Pro is the strongest cheaper option in our database at $0.522 per 1M blended — about 37% less than DeepSeek R1. It is also open-weight, so self-hosting is an option.
Can I run DeepSeek R1 locally?
Yes. DeepSeek R1 is open-weight and needs about ~400 GB of VRAM at 4-bit quantisation (Multi-GPU server).
Why does DeepSeek R1 charge more for output than input?
Output tokens are generated one at a time and cannot be batched the way a prompt can, so they cost the provider more to serve. DeepSeek R1 charges 4.3× more for output, which is why prompt-heavy workloads are far cheaper to run than generation-heavy ones.
See every DeepSeek model priced side by side: DeepSeek API pricing.
Prices are the published list rates for the model's primary API and are reviewed as providers change them. Volume, batch and cached-input discounts are not included. Compare every model side by side in the Datenbank für KI-Modelle oder das LLM-Leaderboard.
