DeepSeek V3.2

DeepSeek V3

Ranks #17 of 43 on reading real contracts into structured billing data — -5.2 pts vs GPT-5.5.

64.1%

Accuracy

#17 of 43

28.3%

Hallucination

HIGH-confidence & wrong

$7.80

Cost / 1k contracts

$0.01 each · via OpenRouter

51.5s

Median latency

p90 57.4s

±5.2

Run-to-run σ

3 runs

8.26

Value

acc. pts per $/1k

Tokens per contract

Input23,075

Output3,893

Reasoning0

Reliability 100% · valid structured output across 18 calls.

Want the full picture? How we score →