ScaleDown CDMs tested: cheap classify, extract and summarize models

A developer benchmarked ScaleDown's Constrained Decoding Models on public data: CLINC150 classification hit 97.3% versus 90.7% for a local Qwen3 1.7B, and receipt extraction reached 89.5% field accuracy with 100% valid JSON. Pricing is $0.05 per million input tokens with free output; the vendor's "467x cheaper" and DecisionBench claims remain unverified.

• CLINC150 classification: 97.3% for ScaleDown vs 90.7% for Qwen3 1.7B
• SROIE receipt extraction: 89.5% field accuracy, 100% valid JSON
• Pricing at $0.05 per million input tokens, output free
• p50 latency 162 ms and p95 397 ms for Decisions