Model Tracker

DeepSeek-R1

DeepSeek-R1 is an AI model made by DeepSeek AI. DeepSeek AI released it on 20 January 2025. It is generally available. It takes text and returns text.

DeepSeek AI

DeepSeek-R1 is an AI model made by DeepSeek AI. DeepSeek AI released it on 20 January 2025. It is generally available. It takes text and returns text. Its context window is 128,000 tokens. Its weights are open. Its licence is MIT License.

Released: 20 January 2025 — deepseek.com, read 16 October 2026
Status: Generally available — github.com, read 16 October 2026
Takes: Text — huggingface.co, read 16 October 2026
Returns: Text — huggingface.co, read 16 October 2026
Open weights: Yes — deepseek.com, read 16 October 2026
Licence: MIT License — github.com, read 16 October 2026
Context window: 128,000 tokens — github.com, read 16 October 2026
Longest answer: Not confirmed: The maker's published 32,768-token maximum is stated for benchmark generation, not as the standard output limit; no standard limit for this model is confirmed on the cited maker pages.
Knowledge cutoff: Not confirmed: The cited maker pages do not state a knowledge-cutoff date for DeepSeek-R1.
API price, input: Not confirmed: The current maker pricing page does not list DeepSeek-R1, and the maker states that the legacy API names were discontinued after 2026-07-24. The R1 release announcement's historical prices are not a current standard-tier price for this model.
API price, output: Not confirmed: The current maker pricing page does not list DeepSeek-R1, and the maker states that the legacy API names were discontinued after 2026-07-24. The R1 release announcement's historical prices are not a current standard-tier price for this model.
Where: Not confirmed: two readings disagree: weights and api, app, weights
Parameters: 671,000,000,000 — github.com, read 16 October 2026
Model card: https://huggingface.co/deepseek-ai/DeepSeek-R1 — huggingface.co, read 16 October 2026

Benchmarks

Measured by others, not by the maker: each score with the name the measurer gives the model and the setting it ran, its day and its licence.

Arena

Text Arena, deepseek-r1: 1,398 (1,393–1,403, 18,524 votes), published 15 October 2026

Arena’s leaderboard dataset (lmarena-ai/leaderboard-dataset on Hugging Face), CC BY 4.0: ratings from people’s votes between two anonymous models, as Arena publishes them. Data · CC BY 4.0

Epoch AI

Epoch Capabilities Index, DeepSeek-R1 (May 2025): 141.3 (138.7–142.8), as of 19 October 2026
Epoch Capabilities Index, DeepSeek-R1: 139.0 (136.1–140.4), as of 19 October 2026
GPQA Diamond, DeepSeek-R1-0528: 76.3% (± 2.4), run 29 May 2025
GPQA Diamond, DeepSeek-R1: 71.7% (± 3.1), run 26 May 2025
OTIS Mock AIME 2024–2025, DeepSeek-R1-0528: 66.4% (± 4.9), run 29 May 2025
OTIS Mock AIME 2024–2025, DeepSeek-R1: 53.3% (± 7.5), run 26 February 2025

Epoch AI, ‘Capabilities & benchmarking’, epoch.ai/benchmarks, CC BY 4.0: Epoch AI’s own runs and its Capabilities Index. Data · CC BY 4.0

Questions

When was DeepSeek-R1 released?

DeepSeek AI released DeepSeek-R1 on 20 January 2025.

Is DeepSeek-R1 still available?

Yes: it is generally available, as DeepSeek AI states it.

What is the context window of DeepSeek-R1?

The context window of DeepSeek-R1 is 128,000 tokens, as DeepSeek AI prints it.

Are the weights of DeepSeek-R1 open?

Yes: DeepSeek AI publishes the weights of DeepSeek-R1 for download.

What is the licence of DeepSeek-R1?

The licence of DeepSeek-R1 is MIT License, as DeepSeek AI names it.

What does DeepSeek-R1 take and return?

DeepSeek-R1 takes text and returns text.

How many parameters does DeepSeek-R1 have?

DeepSeek-R1 has 671,000,000,000 parameters, as DeepSeek AI publishes it.

Sources

DeepSeek-R1 (read 16 October 2026)
DeepSeek-R1 Release (read 16 October 2026)
LICENSE (read 16 October 2026)
deepseek-ai/DeepSeek-R1 (read 16 October 2026)
DeepSeek V4 Preview Release (read 16 October 2026)

All of the Model Tracker