R1

Tool callingReasoning

Hugging Face Launch post Website Docs

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

Context: 64K
Released: 1y ago
Parameters: 1.5B
Input / 1M tokens: $0.70
Output / 1M tokens: $2.50

Sources:models.devOpenRouter

Benchmarks

DeepSWE is the primary coding benchmark where an exact model evaluation exists. Other sourced evaluations appear below; missing scores are not zero.

Not evaluated in DeepSWE 1.1 as of 2026-09-22. Results are only shown for exact model versions in the official dataset. Official leaderboard.

Other published benchmarks

Evaluation	Reported score and source
ARC-AGI-2	1.3% Third-party evaluator · DeepSeek-R1 External evaluation via Epoch AI Original-source terms apply
GPQA diamond	71.7% Epoch AI · DeepSeek-R1 Epoch AI evaluation
OTIS Mock AIME 2024-2025	53.3% Epoch AI · DeepSeek-R1 Epoch AI evaluation

Benchmark compilation: Epoch AI, Capabilities & benchmarking (CC BY attribution). Scores are converted to percentages and model names matched to the Vertical Tabs catalog. Individual evaluations are credited to their original sources. External results may carry separate licences and evaluation configurations; do not interpret scores from different harnesses as directly comparable.

Specifications

Developer: DeepSeek
Model family: DeepSeek R
Type: Language model
Predecessor: —
Successor: —
Built on: DeepSeek-V3based on
Models built on this: R1 0528variant of
DeepSeek-R1-Distill-Qwen-32Bdistilled from
Released: January 20, 2025
Model ID: deepseek/deepseek-r1
Openness: Open weights
License: —
Parameters: 1.5B
Input modalities: text
Output modalities: text
Modality: text->text
Native context: 160K
Tokenizer: DeepSeek
Max output tokens: 16K
Knowledge cutoff: 2024-07
Hugging Face: deepseek-ai/DeepSeek-R1
Source freshness: Metadata synced Oct 8, 2026, 10:17 PM UTC

Open-weight builds

Ollama variants

Every published Ollama tag matched to this model, including parameter editions, instruction/base variants, and quantized builds.

27 builds

Variant	Parameters	Type	Quantization	Size	Context	Input
deepseek-r1:1.5b-qwen-distill-fp16	1.5B	—	FP16	3.6 GB	128K	text
deepseek-r1:1.5b	1.5B	—	Q4_K_M	1.1 GB	128K	text
deepseek-r1:1.5b-qwen-distill-q8_0	1.5B	—	Q8_0	1.9 GB	128K	text
deepseek-r1:7b-qwen-distill-fp16	7B	—	FP16	15 GB	128K	text
deepseek-r1:7b	7B	—	Q4_K_M	4.7 GB	128K	text
deepseek-r1:7b-qwen-distill-q8_0	7B	—	Q8_0	8.1 GB	128K	text
deepseek-r1:8b-0528-qwen3-fp16	8B	—	FP16	16 GB	128K	text
deepseek-r1:8b-llama-distill-fp16	8B	—	FP16	16 GB	128K	text
deepseek-r1:8b-llama-distill-q4_K_M	8B	—	Q4_K_M	4.9 GB	128K	text
deepseek-r1:latest	8B	—	Q4_K_M	5.2 GB	128K	text
deepseek-r1:8b-0528-qwen3-q8_0	8B	—	Q8_0	8.9 GB	128K	text
deepseek-r1:8b-llama-distill-q8_0	8B	—	Q8_0	8.5 GB	128K	text
deepseek-r1:14b-qwen-distill-fp16	14B	—	FP16	30 GB	128K	text
deepseek-r1:14b	14B	—	Q4_K_M	9 GB	128K	text
deepseek-r1:14b-qwen-distill-q8_0	14B	—	Q8_0	16 GB	128K	text
deepseek-r1:32b-qwen-distill-fp16	32B	—	FP16	66 GB	128K	text
deepseek-r1:32b	32B	—	Q4_K_M	20 GB	128K	text
deepseek-r1:32b-qwen-distill-q8_0	32B	—	Q8_0	35 GB	128K	text
deepseek-r1:70b-llama-distill-fp16	70B	—	FP16	141 GB	128K	text
deepseek-r1:70b	70B	—	Q4_K_M	43 GB	128K	text
deepseek-r1:70b-llama-distill-q8_0	70B	—	Q8_0	75 GB	128K	text
deepseek-r1:671b-0528-fp16	671B	—	FP16	1.3 TB	160K	text
deepseek-r1:671b-fp16	671B	—	FP16	1.3 TB	160K	text
deepseek-r1:671b	671B	—	Q4_K_M	404 GB	160K	text
deepseek-r1:671b-q4_K_M	671B	—	Q4_K_M	404 GB	160K	text
deepseek-r1:671b-0528-q8_0	671B	—	Q8_0	713 GB	160K	text
deepseek-r1:671b-q8_0	671B	—	Q8_0	713 GB	160K	text

Additional source evidence

Source model: deepseek/deepseek-r1
Source family: deepseek-thinking
Last updated: 2025-05-29
Knowledge cutoff: 2024-07
Model context: 128K
Model output limit: 33K
Open weights: Yes
File attachments: No
Reasoning: Yes
Tool calling: Yes
Temperature control: Yes

Downloadable weights

Hugging Face

Published benchmarks

Benchmark	Score	Evaluation details	Evidence
Aider Polyglot	56.9 · percent correct	2025-01-20	Source
Artificial Analysis Coding Index	15.9 · index	2026-03-11	Source
SciCode	35.7 · percent correct	2026-03-11	Source
Terminal-Bench Hard	6.1 · success rate	2026-03-11	Source

Complete source record (JSON)

{
  "id": "deepseek/deepseek-r1",
  "name": "DeepSeek-R1",
  "description": "Classic open reasoning model for transparent math, coding, and deliberate problem solving",
  "family": "deepseek-thinking",
  "last_updated": "2025-05-29",
  "knowledge": "2024-07",
  "attachment": false,
  "reasoning": true,
  "tool_call": true,
  "temperature": true,
  "modalities": {
    "input": [
      "text"
    ],
    "output": [
      "text"
    ]
  },
  "limit": {
    "context": 128000,
    "output": 32768
  },
  "benchmarks": [
    {
      "name": "Aider Polyglot",
      "score": 56.9,
      "source": "https://aider.chat/docs/leaderboards/",
      "metric": "percent correct",
      "date": "2025-01-20"
    },
    {
      "name": "Artificial Analysis Coding Index",
      "score": 15.9,
      "source": "https://openrouter.ai/deepseek/deepseek-r1/benchmarks",
      "metric": "index",
      "date": "2026-03-11"
    },
    {
      "name": "SciCode",
      "score": 35.7,
      "source": "https://openrouter.ai/deepseek/deepseek-r1/benchmarks",
      "metric": "percent correct",
      "date": "2026-03-11"
    },
    {
      "name": "Terminal-Bench Hard",
      "score": 6.1,
      "source": "https://openrouter.ai/deepseek/deepseek-r1/benchmarks",
      "metric": "success rate",
      "date": "2026-03-11"
    }
  ],
  "release_date": "2025-01-20",
  "open_weights": true,
  "weights": [
    {
      "label": "Hugging Face",
      "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1"
    }
  ]
}

Infrastructure

Provider availability

All providers

Provider	Endpoint	Context	Input /M tokens	Output /M tokens	Cached input /M tokens	Cache write /M tokens	Uptime 1d	Data handling
Novita	novita/fp8	64K	$0.70	$2.50	—	—	100.00%	ZDR: Yes Training: Unknown Prompt logging: Unknown Retention: Unknown ZDR source Privacy policy Checked 2026-10-15

Provider data:OpenRouter

Deployment graph

Can I run this?

Vertical Tabs only shows compatibility when it has structured evidence tying this exact model to hardware, runtime, quantization, memory use, and observed performance.

Browse hardware