DeepSeek V4 Pro

NewDeepSeek V4

Ranks #26 of 43 on reading real contracts into structured billing data — -9.8 pts vs GPT-5.5.

59.5%

Accuracy

#26 of 43

23.7%

Hallucination

HIGH-confidence & wrong

$45.0

Cost / 1k contracts

$0.04 each · via OpenRouter

143.6s

Median latency

p90 295.4s

±15.1

Run-to-run σ

3 runs

1.32

Value

acc. pts per $/1k

Tokens per contract

Input22,631

Output7,933

Reasoning4,660

Reliability 94.4% · valid structured output across 18 calls.

Want the full picture? How we score →