DeepSeek · open · 2025-05

DeepSeek-R1-0528

Item: DeepSeek-R1-0528
Rating: 74.4

#13 index 74.4

the take
The model that proved open reasoning could be genuinely good. A year on it is verbose, slow, and beaten by its own successors. Historical importance, museum-piece performance.

benchmarks

57.6

81.0

85.0

87.5

LiveCodeBench (coding)

73.3

specs

Context: 128K
Input: $0.55/M
Output: $2.19/M
Speed: —
Modality: text

strengths

Landmark fully-open reasoning model
Strong math and step-by-step reasoning
Widely hosted, cheap, spawned many distills

weaknesses

Over a year old, surpassed by V3.2/V4
Verbose, high latency
Endpoint deprecating 2026-07-24

Sources: [1][2]

Back to leaderboard Battle DeepSeek-R1-0528