Back to models

DeepSeek-R1 (May 2025)

ECI#101
141
DeepSeek
May 28, 2025
Open weights

DeepSeek-R1 (May 2025) is an open-weights model from DeepSeek, released May 28, 2025. It has an ECI score of 141 and ranks 101st of 249 tracked models. At the time of its release, it ranked 9th of 140 tracked models. At $0.70/Mtok for input tokens and $2.40/Mtok for output tokens, it is less expensive per token than similarly-capable models.

Input price$0.70 / 1M tokens
Output price$2.40 / 1M tokens
Parameters6.7e11
Training compute4.0e24 FLOP

DeepSeek-R1 (May 2025) vs select alternatives

Reasoning
DomainBenchmark
DeepSeek-R1 (May 2025)
DeepSeek
May 28, 2025
Open weights
DeepSeek-R1
DeepSeek
Jan. 20, 2025
Open weights
Gemini 2.5 Flash (May 2025)
Google DeepMind
May 20, 2025
Closed weights
Claude Sonnet 4
Anthropic
May 22, 2025
Closed weights
ECI scoreBest score 166
141139142142
Ranking#101#116#100#99
Conceptual reasoning
SimpleBenchBest score 81.9%
40.8%
30.9%
n/a
45.5%
ARC-AGI-2Best score 95.0%
1.1%
1.3%
2.5%
5.9%
ARC-AGI-1Best score 98.5%
21.2%
15.8%
33.3%
40.0%
Mathematics
MATH Level 5Best score 98.1%
96.6%
93.1%
n/a
84.4%
OTIS Mock AIME 2024-2025Best score 100.0%
66.4%
53.3%
70.8%
71.1%
Science
GPQA DiamondBest score 95.8%
76.3%
69.2%
n/a
78.3%
Software engineering
METR Time HorizonsBest score 1044.8
31.226.9—74.9
Aider PolyglotBest score 88.0%
71.4%
56.9%
55.1%
61.3%
WeirdML (v2)Best score 92.9%
41.6%
36.5%
40.9%
46.1%

Model metadata

General
OrganizationDeepSeek
Release dateMay 28, 2025
Model accessibilityOpen weights
CountryChina
Knowledge cutoffUnknown
Provider notesDeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning
Pricing
Input price$0.70 / 1M tokens
Output price$2.40 / 1M tokens
Training
Parameters6.7e11
Training compute4.0e24 FLOP
Training hardwareUnknown
Training timeUnknown
Training dataset size1.5e13 datapoints