Model
DeepSeek-R1-0528 NO.31
Iconic open reasoning model — strong math at low cost, now a generation behind.
DeepSeek · DeepSeek · Reasoning · Open weights
Specification
13 fields
Fig.01 — DeepSeek — DeepSeek
CHINA
- Context window
- 164Ktokens
- Max output
- 66Ktokens
- Input price
- $0.50/ 1M
- Output price
- $2.15/ 1M
- Cached input
- $0.14/ 1M
- Throughput
- —
- Class
- Reasoning
- Modalities
- Text
- Weights
- Open
- Parameters
- 671B
- License
- MIT
- Knowledge cutoff
- Apr 2025
- Released
- May 2025
Overview
DeepSeek-R1-0528 is DeepSeek's open-weight reasoning flagship (671B MoE, 37B active, MIT), known for transparent long chain-of-thought and excellent math and competition performance at a fraction of closed-model cost. It remains a popular self-hosted reasoner, though 2026 flagships have surpassed it and its long reasoning traces make it slow. Best for math, logic, and reasoning-heavy tasks rather than long agentic coding.
Strengths
- Elite math and competition reasoning
- Transparent, fully open MIT-licensed weights
- Strong on GPQA-Diamond and AIME-style problems
- Distillable into smaller models for cheaper reasoning
Trade-offs
- Slow and token-heavy due to long reasoning traces
- Agentic-coding reliability below the frontier
- Surpassed by newer 2026 reasoners
- Text-only, verbose on simple queries
Fit
Best for
- Hard math and scientific reasoning
- Competitive programming and algorithm design
- Reasoning research and distillation
- Self-hosted reasoning workloads
Not ideal for
- Low-latency interactive chat
- Long autonomous agentic coding
Capabilities
Normalized 0—100
Ten axes, normalized 0–100 and scored the same way across the whole catalog — so a 78 here means what a 78 means anywhere else on the bench.
Fig.02 — Capability radar
0 — 100
- Reasoning
- 85
- Coding
- 68
- Math
- 92
- Writing
- 74
- Knowledge
- 82
- Speed
- 45
- Agentic
- 62
- Vision
- 5
- Multilingual
- 80
- Long Context
- 74
Benchmarks
08 results
Public results, with independent third-party runs marked. Bars normalize percentages against 100 and Elo ratings against a 1500 ceiling.
- Aider PolyglotIndependentCoding · Aider
- 71.6%
- MATHMath · Vendor-reported
- 97.3%
- MMLU-ProKnowledge · Vendor-reported
- 85%
- GPQA DiamondIndependentReasoning · Independent aggregators
- 81%
- AIME 2025Math · Vendor-reported
- 87.5%
- LiveCodeBenchIndependentCoding · LiveCodeBench
- 65.9%
- SWE-bench VerifiedIndependentCoding · SWE-bench
- 57.6%
- LMArena EloIndependentGeneral · LMArena
- 1410Elo
Alternatives
03 comparable
Models in roughly the same class — the ones worth weighing against this record.
Index
Record 31 of 41