Cost mode:

Category: Structured Data & Fact Extraction · Rail: absolute · Typical I/O: 1098→1810 tokens

Models

Frontier on this task: Moonshot Kimi K3 at 8.84 / 10. Quality bar at 90%: 7.95.

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
Claude Sonnet 58.77 / 108.59$4.57best value
Thinking Machines Inkling Small8.50 / 108.29$5.261.2x more expensive
Thinking Machines Inkling8.39 / 108.17$10.182.2x more expensive
Moonshot Kimi K38.84 / 108.63$18.844.1x more expensive
Gemini 3.5 Flash Lite5.32 / 104.97$0.4091% cheaper
Gemini 3.5 Flash7.27 / 106.98$3.5223% cheaper

Cost breakdown

ModelQualityConfidenceCost / 1k runsOverpayMode
Claude Sonnet 5 Anthropic8.77 / 10 CI [8.59, 8.95]RANKED$4.57best valuebatch
Thinking Machines Inkling Small OpenRouter8.50 / 10 CI [8.29, 8.72]HIGH$5.261.2xbatch
Thinking Machines Inkling OpenRouter8.39 / 10 CI [8.17, 8.60]HIGH$10.182.2xbatch
Moonshot Kimi K3 best Moonshot AI8.84 / 10 CI [8.63, 9.04]HIGH$18.844.1xbatch

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: 1098 input tokens → 1810 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.

Evaluation rubric

Judge evidence-based regional relevance, ranking, coverage versus focus trade-off, allowed-set compliance, and restraint against unsupported regions. Cardinality and identifier validity are deterministic.

Prompt templates

The system + user template pair used for this task.

RESEARCH_REGION_IDENTIFIER_SYSTEM + RESEARCH_REGION_IDENTIFIER_USER (286 calls in window)

System prompt

You are an expert market analyst specializing in geographic market identification.

Your task is to identify the most relevant geographic regions for researching a specific subject.

Consider:
1. Where is the subject primarily headquartered or based?
2. What are the subject's primary markets?
3. Where does the subject have significant operations?
4. Where would authoritative information be published?
5. What regulatory jurisdictions are most relevant?

Output a JSON object with:
{{
  "region_codes": ["US", "UK", ...],
  "reasoning": "Brief explanation of why these regions were selected"
}}

Select 1-4 most relevant regions. Prioritize quality over quantity.

User prompt

Identify the most relevant geographic regions for researching: {subject_name}

Subject Code: {subject_code}
Subject Type: {subject_type}
Subject Description: {subject_description}

Available Regions:
{available_regions}

Additional Context:
{additional_context}