readable AI benchmarks (simplified)

readable AI benchmarks

DeepSeek

DeepSeek V3.2 Speciale

reasoning · open weights · Dec 1, 2025

DeepSeek V3.2 Speciale has a Quality Score of 31.7, ranking 238th among 551 scored models. It has a 128k token context window. Its Reliability Score is 41.1, ranking 207th among 553 scored models, above average (total average: 35.6).

Quality Score31.7coverage 100.0%: AA Intelligence and AA-Omniscience outcomes available
Value Scorenot available
Factual reliability41.1
Cache Discountnot available

Model specification

Reasoning
reasoning
Input modalities
none
Output modalities
none
Context window
128k tokens
Weights
open weights

Token prices USD per 1M tokens

Input
not available
Output
not available
Cached Input
not available
Cache write
not available
Cost per task
not available

Capability

Intelligence
14.5
Coding
not available
Agentic
not available
Omniscience
-17.7
Correct
37.8%
Blended price
not available

Answer outcomes

Fully graded outcomes (Correct + Incorrect): 93.2%

Correct 37.8%Incorrect 55.5%Partial / not attempted 6.8%

Artificial Analysis benchmarks

Humanity’s Last Exam
29%
GPQA Diamond
87%
CritPt
7%
AA-Omniscience
41%
AA-LCR
70%

Similar models

Support me! Patreon