DeepSeek

DeepSeek-R1

Generally availableOpen weights64K contextReasoning

Summary

DeepSeek-R1 is an open-weight language model from DeepSeek, released on 20 Jan 2025. It accepts text and generates text, with a 64K-token context window and up to 33K output tokens per response. As of 18 Nov 2026, the lowest listed API price is $0.70 per million input tokens and $2.50 per million output tokens (Hugging Face Inference Providers), across 3 providers we track.

Classic open reasoning model for transparent math, coding, and deliberate problem solving

Last verified 18 Nov 2026. Every figure on this page links to its source.

Specifications

DeveloperDeepSeek
Release date20 Jan 2025 [models.dev]
StatusGenerally available [models.dev]
WeightsOpen weights [Hugging Face]
Licensemit [Hugging Face]
Parameters684B [Hugging Face]
Context window64K (64,000 tokens) [models.dev]
Max output33K tokens [models.dev]
Knowledge cutoffJul 2024 [models.dev]
InputText [models.dev]
OutputText [models.dev]
Reasoning modeYes [models.dev]
Tool callingYes [models.dev]
Structured outputYes [models.dev]
Hugging Facedeepseek-ai/DeepSeek-R1 [Hugging Face]

DeepSeek-R1 API pricing by provider

USD per million tokens. Sorted by input price.

ProviderInputOutputCached inputContextSource
Hugging Face Inference Providers
deepseek-ai/DeepSeek-R1
$0.70$2.50–64Kmodels.dev, 18 Nov 2026
Novita AI
deepseek/deepseek-r1-turbo
$0.70$2.50–64Kmodels.dev, 18 Nov 2026
Snowflake Cortex
deepseek-r1
–––128Kmodels.dev, 18 Nov 2026

Running DeepSeek-R1 locally

Estimated memory, from the parameter count and the architecture in config.json on Hugging Face. Includes about 0.8 GB of runtime overhead.

QuantizationWeightsTotal, 8K contextTotal, 32K context
Q4_K_M (4-bit)415 GB416 GB418 GB
Q8_0 (8-bit)727 GB729 GB730 GB
FP16 / BF16 (16-bit)1369 GB1370 GB1372 GB

At 4-bit with an 8K conversation, DeepSeek-R1 needs about 416 GB, more than any single consumer graphics card; it needs a unified-memory machine, a multi-GPU setup or a split between graphics card and system RAM. Check your own hardware, or see Best computer for local AI on our sister site AIPCs.

Frequently asked questions

What is the context window of DeepSeek-R1?

DeepSeek-R1 has a context window of 64,000 tokens (64K) and can generate up to 32,768 tokens in a single response, according to models.dev.

How much does the DeepSeek-R1 API cost?

DeepSeek-R1 is listed from $0.70 per million input tokens and $2.50 per million output tokens (Hugging Face Inference Providers). Prices last verified 18 Nov 2026.

Is DeepSeek-R1 open source?

DeepSeek-R1 is an open-weight model: its weights are published on Hugging Face as deepseek-ai/DeepSeek-R1 under the mit license. Check the license terms before commercial use.

When was DeepSeek-R1 released?

DeepSeek released DeepSeek-R1 on 20 Jan 2025, according to models.dev.

Which providers offer DeepSeek-R1?

We track DeepSeek-R1 on 3 providers: Hugging Face Inference Providers, Novita AI, Snowflake Cortex.

How much VRAM does DeepSeek-R1 need?

About 416 GB at 4-bit (Q4_K_M) with an 8K-token conversation, and about 729 GB at 8-bit. That is more than any single consumer graphics card holds. These are estimates from the model's published parameter count and architecture.

Similar models from other labs

Comparable on output type, weights, release date, price and context window. Prices are the lowest listed, USD per million tokens.

ModelLabReleasedContextInput $/1MOutput $/1MWeights
Aion-RP 1.0 (8B)Aion Labs4 Feb 202533K$0.80$1.60Open
Qwen2.5-VL 72B InstructAlibaba (Qwen)26 Jan 2025128K$0.80$1Open
GLM-4.5VZhipu AI (Z.ai)11 Aug 202564K$0.60$1.80Open
Aya Expanse 32BCohere24 Oct 2024128K$0.50$1.50Open
Skyfall 36B V2Thedrummer10 Mar 202533K$0.55$0.80Open
Kimi K2 InstructMoonshot AI11 Jul 2025131K$0.57$2.30Open

Change history

  • Max output changed from 16K to 33K
  • Name changed from DeepSeek R1 (Turbo) to DeepSeek-R1
  • Name changed from DeepSeek-R1 to DeepSeek R1 (Turbo)
  • Max output changed from 33K to 16K
  • Max output changed from 16K to 33K
  • Name changed from DeepSeek R1 (Turbo) to DeepSeek-R1
  • Max output changed from 33K to 16K
  • Name changed from DeepSeek-R1 to DeepSeek R1 (Turbo)
  • No longer listed on cloudflare-workers-ai
  • Name changed from Deepseek R1 Distill Qwen 32B to DeepSeek-R1
  • Structured output changed from no to yes
  • Context window changed from 80K to 64K
  • Max output changed from 80K to 33K
  • Tool calling changed from no to yes
  • Name changed from DeepSeek R1 (Turbo) to Deepseek R1 Distill Qwen 32B
  • Max output changed from 16K to 80K
  • Tool calling changed from yes to no
  • Context window changed from 64K to 80K
  • Structured output changed from yes to no
  • Max output changed from 33K to 16K
  • Name changed from DeepSeek-R1 to DeepSeek R1 (Turbo)
  • Name changed from Deepseek R1 Distill Qwen 32B to DeepSeek-R1
  • Context window changed from 80K to 64K
  • Structured output changed from no to yes
  • Max output changed from 80K to 33K

Sources