DeepSeek V4 Flash

NewDeepSeek V4

Ranks #11 of 43 on reading real contracts into structured billing data — -1.7 pts vs GPT-5.5.

67.6%

Accuracy

#11 of 43

23%

Hallucination

HIGH-confidence & wrong

$5.10

Cost / 1k contracts

$0.01 each · via OpenRouter

102.9s

Median latency

p90 144.6s

±7.4

Run-to-run σ

3 runs

13.15

Value

acc. pts per $/1k

Tokens per contract

Input22,882

Output6,917

Reasoning3,890

Reliability 83.3% · valid structured output across 18 calls.

Want the full picture? How we score →