GLM-5.2

NewGLM

Ranks #40 of 43 on reading real contracts into structured billing data — -29.7 pts vs GPT-5.5.

39.6%

Accuracy

#40 of 43

48.6%

Hallucination

HIGH-confidence & wrong

$29.1

Cost / 1k contracts

$0.03 each · via OpenRouter

83.2s

Median latency

p90 181.2s

±9.9

Run-to-run σ

3 runs

1.36

Value

acc. pts per $/1k

Tokens per contract

Input21,623

Output7,837

Reasoning4,871

Reliability 83.3% · valid structured output across 18 calls.

Want the full picture? How we score →