DeepSeek R1
Jan 2025
DeepSeek R1Compare inon
Benchmarks of DeepSeek R120
BenchmarkByScoreRank
Overall1
Knowledge4
Science3
Coding3
Agents3
Documents2
Legacy · saturated or no longer reported4
No benchmark matches
Ranking?
Alternatives?
ModelScoreCostPer taskWhat changes
CheaperGPT-6.1 Solmedium+72.2-18%$0.21Remove unsupported parameters+5BetterGPT-6 Lunalow+23.1-98%$0.005Function calling only with none · Chat Completions+3
OpenMuse Spark Openxhigh+56.9+223%$0.84Memory (4-bit): 1.2 GB → 493 GB+2CheaperGPT-6.1 Terraxhigh+60.9-26%$0.19CheaperGPT-6 Solmedium+59.2-6%$0.25ModelScoreCostPer taskWhat changes
GPT-6.1 Solmedium+72.2-18%$0.21Remove unsupported parameters+5GPT-6.1 Terraxhigh+60.9-26%$0.19GPT-6 Solmedium+59.2-6%$0.25
Grok 4.3high+50.2-22%$0.21
MiniMax M3.1 Flashthinking+47.8-16%$0.22No strict JSON+1ModelScoreCostPer taskWhat changes
GPT-6 Lunalow+23.1-98%$0.005Function calling only with none · Chat Completions+3
Granite 4.2 3BthinkingOpen+17.5-98%$0.006Memory (4-bit): 1.2 GB → 1.8 GB
Gemma 4 E4BthinkingOpen+12.5-97%$0.008Memory (4-bit): 1.2 GB → 4 GBGPT-5.6 Lunalow+17.6-96%$0.01Granite 4.2 8BthinkingOpen+15.1-91%$0.024Memory (4-bit): 1.2 GB → 4.4 GBModelScoreCostPer taskWhat changes
Muse Spark OpenxhighOpen+56.9+223%$0.84Memory (4-bit): 1.2 GB → 493 GB+2
Kimi K3maxOpen+51.9+664%$2.00Memory (4-bit): 1.2 GB → 1,668 GB+4
Step 5 PreviewthinkingOpen+48.6+292%$1.03Memory (4-bit): 1.2 GB → 363 GB+3
GLM 5.3maxOpen+46.5+666%$2.01Memory (4-bit): 1.2 GB → 452 GB+4MMotif 3Open+42.4-22%$0.50 / $2Memory (4-bit): 1.2 GB → 189 GBModelScoreCostPer taskWhat changes
DeepSeek V4.1 Flash 1017maxOpen+36.9same$0.26Memory (4-bit): 1.2 GB → 458 GB+5DeepSeek V4 Pro 0813non-thinkingOpen+19.8+36%$0.36Memory (4-bit): 1.2 GB → 990 GB+5DeepSeek V3.2 SpecialethinkingOpen+14.5——Memory (4-bit): 1.2 GB → 411 GBR1 0528thinkingOpen+4.8—$0.39 / $1.63Memory (4-bit): 1.2 GB → 411 GBDeepSeek V3 0324non-thinkingOpen-8.4-92%$0.021Memory (4-bit): 1.2 GB → 411 GBDeepSeek V3non-thinkingOpen-9.4-92%$0.02Memory (4-bit): 1.2 GB → 411 GBDeepSeek V3.1non-thinkingOpen-10.5—$0.25 / $0.95Memory (4-bit): 1.2 GB → 411 GBDeepSeek V3.1 Terminusnon-thinkingOpen-11.3—$0.27 / $0.95Memory (4-bit): 1.2 GB → 411 GBAll DeepSeek models →Run it yourself
estimateFits on✓ 8 GB (phone, laptop) · compressed✓ 16 GB · compressed✓ 24 GB GPU · compressed✓ 48 GB GPU · compressed✓ 80 GB GPU · compressed
LicenceMITfree for commercial use
Benchmarks
all categoriesChat & Knowledge
Humanity's Last Exam8.5#213
AA Omniscience Index-32.2#158
CritPt (physics)0.6#144
AA Intelligence Index 4.3.211.4#194
AA Omniscience · Accuracy30.5#78
AA-LCR (long context) 1.157.7#165
GDP.pdf · AA all-pass2.4#109
GPQA Diamond pass@1lab71.5
MMLU pass@1lab90.8
AIME 2024 pass@1lab79.8
MATH-500 pass@1lab97.3
GPQA Diamond70.8#183
Coding
SciCode38.3#109
SWE-bench Verified resolvedlab49.2
Codeforces percentilelab96.3
Terminal-Bench 2.119.1#114
LiveCodeBench · AA archived61.7#72
Agents & Tool Use
AutomationBench-AA0.2#131
τ³-Banking · AA6.4#107
AA-Briefcase 1.1 · Elo128.1#124
τ²-bench · AA11.4#232
Writing & Language
IFBench (instructions)39#168
FAQ
How much does DeepSeek R1 cost?›
$0.70 per 1M input tokens and $2.40 per 1M output tokens (cheapest host: DeepInfra). A typical benchmark task costs about $0.26 at default effort.
Is DeepSeek R1 free to use?›
There is no free API; the cheapest paid price is $0.70 per 1M input tokens. The weights are open, so you can run it yourself.
How good is DeepSeek R1, and what is it best at?›
#212 of 360 in Chat & Knowledge, #78 of 137 in Coding.
Is DeepSeek R1 better than GLM-4.6?›
On AA Omniscience Index, DeepSeek R1 scores -32.2 vs -31.7 — about the same.
How can I use DeepSeek R1?›
Through the API of DeepInfra, Novita.
What is the context window of DeepSeek R1?›
128K tokens.
When was DeepSeek R1 released? Is there a newer version?›
Released 2025-01-20. It is the current version in its line.
Is DeepSeek R1 open source?›
Yes, the weights are open (MIT licence).
Can I run DeepSeek R1 locally?›
Yes, with about 1 GB of GPU or unified memory.
Can DeepSeek R1 read images?›
No, it accepts text only.
What is DeepSeek R1?›
DeepSeek R1 is an open-weights language model from DeepSeek, released Jan 2025.
What are the best alternatives to DeepSeek R1?›
GPT-6.1 Sol (cheaper), GPT-6 Luna (better), Muse Spark Open (open weights).