Llama 3.3 70B

Llama 3

Ranks #43 of 43 on reading real contracts into structured billing data — -39.1 pts vs GPT-5.5.

30.2%

Accuracy

#43 of 43

25.4%

Hallucination

HIGH-confidence & wrong

$2.80

Cost / 1k contracts

$0.00 each · via OpenRouter

40.1s

Median latency

p90 66.7s

±16.7

Run-to-run σ

3 runs

10.84

Value

acc. pts per $/1k

Tokens per contract

Input21,235

Output2,073

Reasoning0

Reliability 88.9% · valid structured output across 18 calls.

Want the full picture? How we score →