← DeepSeek Individual model view DeepSeek

deepseek-r1

Appears in 6 benchmarksMean lean (confidence − pass rate): +0.094/6 benchmarks lean overconfident (prospective probe)

Positioning spread: every benchmark, one model

0.000.250.500.751.00SQuAD (factual recall)MMLU-Pro (knowledge)LegalBench (legal reasoning)MathBench (competition math)OmniMath (advanced math)SciCode (scientific code)performanceconfidence (red gap = overconfident)
Performance vs. confidence for deepseek-r1, per benchmark (prospective probe).
BenchmarkTask accConfidenceF₁Leans
SQuAD (factual recall)0.570.530.73-0.04 calibrated
MMLU-Pro (knowledge)0.650.800.80+0.16 overconfident
LegalBench (legal reasoning)0.870.820.85-0.05 calibrated
MathBench (competition math)0.981.000.99+0.02 calibrated
OmniMath (advanced math)0.700.970.84+0.27 overconfident
SciCode (scientific code)0.560.750.68+0.20 overconfident

In the full cloud

-3-3-2-2-1-100112233Performance z-score within benchmark/probe →Confidence z-score →
deepseek-r1 conditions all other model/condition points equal relative confidence and pass rate

Pairwise signal: pairs involving deepseek-r1

Match accuracy controls for the performance base-rate gap
deepseek-r1 pairs
18/ 171
deepseek-r1 mean tau
+0.021
All-pairs mean
+0.037
deepseek-r1 p<0.05
5(27.8%)
-1.0-0.50.00.51.0Pair signal: do confidence gaps rank performance gaps? (Kendall tau-b)
all model pairs (observed) base-rate-matched null calibration-preserving null deepseek-r1 pair (filled = p<0.05) deepseek-r1 mean all-pairs mean

The four metacognitive outcomes

No curated cases for this selection yet — outcome-matrix extraction currently covers a sample of MMLU-Pro trials.
vs deepseek-chat → Compare with anything →