Best LLMs for Metadata Paragraph Rewriting
Polishes a draft metadata paragraph (research period, source counts, source types) at the head of a report. Preserves all facts and approximate length; only the wording is rewritten for clarity and professional tone.
Models
Frontier on this task: Claude Sonnet 5 at 9.51 / 10. Quality bar at 90%: 8.56.
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.
| Model | Quality score | CI low | Cost / 1k runs | vs best value |
|---|---|---|---|---|
| Claude Sonnet 5 | 9.51 / 10 | 9.40 | $1.71 | best value |
| Gemini 3.5 Flash | 9.10 / 10 | 8.90 | $5.93 | 3.5x more expensive |
| GPT-5.4 Mini | 6.68 / 10 | 6.30 | $0.41 | 76% cheaper |
| GPT-5.5 | 7.61 / 10 | 7.27 | $4.92 | 2.9x more expensive |
| Claude Haiku 4.5 | 5.61 / 10 | 5.11 | $1.03 | 40% cheaper |
| Gemini 3.1 Flash Lite | 7.58 / 10 | 7.35 | $0.27 | 84% cheaper |
| DeepSeek V4 Pro | 8.18 / 10 | 7.88 | $0.45 | 74% cheaper |
| Qwen 3.6 Plus | 8.50 / 10 | 8.27 | $1.41 | 18% cheaper |
| Qwen 3.5 Flash | 8.07 / 10 | 7.83 | $0.74 | 57% cheaper |
| GPT-5.4 Nano | 7.15 / 10 | 6.80 | $0.22 | 87% cheaper |
| Gemini 3.1 Pro Preview | 7.45 / 10 | 7.19 | $1.75 | 1x more expensive |
| NVIDIA Nemotron-3 Nano 30B-A3B | 6.55 / 10 | 6.17 | $0.13 | 92% cheaper |
| NVIDIA Nemotron-3 Ultra 550B | 8.55 / 10 | 8.25 | $2.23 | 1.3x more expensive |
| GPT-5.6 Sol | 6.09 / 10 | 5.81 | $2.96 | 1.7x more expensive |
| NVIDIA Nemotron-3 Super 120B | 7.35 / 10 | 6.85 | $0.34 | 80% cheaper |
| Qwen 3.7 Plus | 7.18 / 10 | 7.01 | $4.91 | 2.9x more expensive |
| GPT-5.6 Terra | 6.40 / 10 | 6.16 | $1.11 | 35% cheaper |
| Tencent Hy3 | 6.70 / 10 | 6.47 | $0.52 | 69% cheaper |
| GPT-5.6 Luna | 6.34 / 10 | 6.09 | $0.60 | 65% cheaper |
| Meta Muse Spark 1.1 | 5.80 / 10 | 5.54 | $9.62 | 5.6x more expensive |
| Qwen 3.6 Flash | 6.96 / 10 | 6.82 | $5.46 | 3.2x more expensive |
| Grok 4.5 | 6.11 / 10 | 5.92 | $10.72 | 6.3x more expensive |
Cost breakdown
| Model | Quality | Confidence | Cost / 1k runs | Overpay | Mode |
|---|---|---|---|---|---|
| Claude Sonnet 5 ★ best Anthropic | 9.51 / 10 CI [9.40, 9.61] | RANKED | $1.71 | best value | batch |
| Gemini 3.5 Flash Gemini | 9.10 / 10 CI [8.90, 9.31] | HIGH | $5.93 | 3.5x | batch |
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: 409 input tokens → 552 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
This is a pooled capability — 2 prompt families share it. The pair shown first is the most frequently used in production.
METADATA_PARAGRAPH_IMPROVEMENT_SYSTEM_PROMPT +
METADATA_PARAGRAPH_IMPROVEMENT_USER_PROMPT
(1647 calls in window)
System prompt
You are an expert editor specializing in creating polished, professional metadata paragraphs for research reports. Your task is to take a draft metadata paragraph that describes the research period and sources, and improve its wording to make it: - More professional and polished - Clear and concise - Reader-friendly - Accurate to the data provided IMPORTANT GUIDELINES: 1. Preserve all factual information (dates, numbers, source types) 2. Do NOT add information that wasn't in the original 3. Do NOT change the meaning or facts 4. Keep the paragraph approximately the same length 5. Maintain a professional, informative tone 6. Start with a clear statement about the research period 7. Include source count and types naturally in the flow The improved paragraph should be suitable for inclusion at the beginning of a professional research report or analysis document.
User prompt
Please improve the wording of this metadata paragraph while preserving all factual information:
{draft_paragraph}
Key facts to preserve:
- Date range: {date_from} to {date_to}
- Number of sources: {source_count}
- Content types: {content_type_summary}
Return the improved paragraph that maintains these facts while improving clarity and professionalism.JSON_REPAIR_SYSTEM +
JSON_REPAIR_USER
(5 calls in window)
System prompt
You are a JSON repair tool. The user gives you malformed or partial model output and a JSON Schema. Return ONLY a single valid JSON object that satisfies the schema, salvaging as much real content from the input as possible. Do not invent data for fields the input doesn't support — use the schema's allowed empty/null values. Output the JSON object only: no prose, no markdown, no code fences.
User prompt
JSON Schema:
{schema_json}
Malformed output to repair:
{raw_text}
Return only the corrected JSON object.