Cost mode:

Category: Structured Data & Fact Extraction · Rail: absolute · Typical I/O: 3921→5267 tokens

Models

Frontier on this task: Gemini 3.5 Flash at 9.78 / 10. Quality bar at 90%: 8.80.

point-estimate floor (CI low) · upper CI (less certain) · Bars sorted by blended cost; best-value model first. Greyed rows are MEDIUM+ models whose point estimate clears the bar but whose CI low does not.

ModelQuality scoreCI lowCost / 1k runsvs best value
DeepSeek V4 Flash9.54 / 109.40$1.77best value
GPT-5.6 Luna9.67 / 109.50$2.751.6x more expensive
MiniMax M39.36 / 109.21$6.233.5x more expensive
GPT-5.4 Nano9.46 / 109.22$8.224.6x more expensive
Qwen 3.7 Plus9.67 / 109.52$23.5913x more expensive
GPT-5.6 Terra8.84 / 108.48$32.3018x more expensive
GPT-5.6 Sol9.74 / 109.60$41.7524x more expensive
Gemini 3.5 Flash9.78 / 109.72$45.3126x more expensive
DeepSeek V4 Pro9.54 / 109.40$51.8729x more expensive
Claude Haiku 4.58.28 / 107.90$31.8818x more expensive

Cost breakdown

ModelQualityConfidenceCost / 1k runsOverpayMode
DeepSeek V4 Flash DeepSeek9.54 / 10 CI [9.40, 9.68]RANKED$1.77best valuebatch
GPT-5.6 Luna OpenAI9.67 / 10 CI [9.50, 9.84]RANKED$2.751.6xbatch
MiniMax M3 OpenRouter9.36 / 10 CI [9.21, 9.52]RANKED$6.233.5xbatch
GPT-5.4 Nano OpenAI9.46 / 10 CI [9.22, 9.71]HIGH$8.224.6xbatch
Qwen 3.7 Plus Alibaba Cloud (DashScope)9.67 / 10 CI [9.52, 9.82]RANKED$23.5913xbatch
GPT-5.6 Terra OpenAI8.84 / 10 CI [8.48, 9.21]MEDIUM$32.3018xbatch
GPT-5.6 Sol OpenAI9.74 / 10 CI [9.60, 9.87]RANKED$41.7524xbatch
Gemini 3.5 Flash best Gemini9.78 / 10 CI [9.72, 9.85]RANKED$45.3126xbatch
DeepSeek V4 Pro DeepSeek9.54 / 10 CI [9.40, 9.69]RANKED$51.8729xbatch

Overpay shows how much more you pay than the best-value model that clears the quality bar (marked ★) — the best-value good-enough option. "16x" means you overpay 16× — 16× that reference for no quality benefit above the bar. Typical call shape for this task: 3921 input tokens → 5267 output tokens, EMA-tracked from production traffic. Cost is the observed, all-in $ per 1,000 task runs: each model's own measured usage on this task — output verbosity, thinking/reasoning tokens, cache reads and writes, and the spend on its billed failures — priced at current list rates and adjusted by the billing overhead we actually reconcile against provider invoices. Models that answer tersely cost what they actually cost; models that think at length pay for it. Not comparable to providers' advertised $/1M list rates — this is what running the task costs, not a per-token price.

Prompt templates

The system + user template pair used for this task.

STRUCTURED_OUTPUT_EXTRACTION_SYSTEM_PROMPT + STRUCTURED_OUTPUT_EXTRACTION_USER_PROMPT (1977 calls in window)

System prompt

You are a JSON extraction assistant. Extract structured data from the provided text that matches the given JSON schema. Output ONLY the structured data matching the schema.

User prompt

Extract the structured data from the following text that matches this JSON schema:

## JSON Schema:
```json
{schema_json}
```

## Text to extract from:
{raw_text}