DeepSeek R1

Ranks #27 of 43 on reading real contracts into structured billing data — -10 pts vs GPT-5.5.

59.3%

Accuracy

#27 of 43

24.8%

Hallucination

HIGH-confidence & wrong

$29.3

Cost / 1k contracts

$0.03 each · via OpenRouter

251s

Median latency

p90 590.9s

±6.7

Run-to-run σ

3 runs

2.03

Value

acc. pts per $/1k

Tokens per contract

Input22,105

Output8,480

Reasoning5,619

Reliability 83.3% · valid structured output across 18 calls.

Want the full picture? How we score →