← Back to directory
DeepSeek
DeepSeek-R1
Released Jan 20, 2025 · knowledge cutoff 2024-07
- Status
- Superseded
- Location
- China
- Type
- LLM
- Modality
- Text
- Context window
- 128K
- Max output
- 33K
- Speed
- —
- Price ($/MTok in / out)
- $0.55 / $2.19
- Cost per task
- $0.25
- Open weights
- Yes
Benchmarks7 of 10 reported
MMLU-Proretired
84%
GPQA Diamond
71.5%
SWE-bench Verified
49.2%
AIME (latest)retired
79.8%
Humanity's Last Exam
8.6%
LMArena Elo
1358
ARC-AGI-2
1.3%
SWE-bench Pro
—
Terminal-Bench 2.1
—
GDPval-AA v2
—
Strengths & weaknesses
Strengths
- Shows its full reasoning trace — the model a lot of people learned reasoning from
- Open weights meant anyone could distil it into smaller reasoning models
- Genuinely strong maths and logic without a frontier lab's budget behind it
Weaknesses
- Reasoning runs long and repetitive, inflating both latency and cost
- Self-censors mid-answer on China-sensitive subjects
- Instruction-following and formatting lag well behind its raw reasoning
The 'DeepSeek moment' of January 2025.
For developers
API model strings
Not researched
Licence
MIT License · Permissive · commercial use permitted
Retirement
No retirement announced
Lineage
No recorded predecessor