Best LLMs for Structured Content Summarization
Produces a comprehensive, source-grounded summary of supplied content, optionally organized around a caller-provided structure or analytical concerns. It can isolate the substantive material in noisy web-page text containing navigation, formatting remnants, advertisements, consen
Run this task on Fronset — request an invitation
Models
Frontier on this task: Meta Muse Spark 1.3 at 8.97 / 10. Quality bar at 90%: 8.08.
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 does not clear the bar.
| Model | Quality score | CI low | Cost / 1k runs | vs best value |
|---|---|---|---|---|
| GPT-6 Luna | 8.54 / 10 | 8.32 | $0.41 | best value |
| Claude Haiku 5.5 | 8.16 / 10 | 7.87 | $0.50 | 1.2x more expensive |
| MiniMax M3 | 8.38 / 10 | 8.25 | $1.53 | 3.7x more expensive |
| DeepSeek V4.1 Flash | 8.13 / 10 | 7.81 | $1.57 | 3.8x more expensive |
| Tencent Hy3 | 8.69 / 10 | 8.52 | $3.44 | 8.3x more expensive |
| Thinking Machines Inkling Small | 8.32 / 10 | 8.01 | $4.50 | 11x more expensive |
| GLM-5.3 Flash | 8.46 / 10 | 8.28 | $5.66 | 14x more expensive |
| GPT-6 Sol | 8.24 / 10 | 7.96 | $6.67 | 16x more expensive |
| Claude Opus 5.5 | 8.14 / 10 | 7.82 | $21.94 | 53x more expensive |
| Grok 4.7 | 8.52 / 10 | 8.25 | $26.80 | 65x more expensive |
| Meta Muse Spark 1.3 | 8.97 / 10 | 8.82 | $27.69 | 67x more expensive |
| Thinking Machines Inkling | 8.29 / 10 | 8.01 | $37.32 | 90x more expensive |
| Moonshot Kimi K3 | 8.68 / 10 | 8.47 | $40.24 | 97x more expensive |
| Tencent Hy4 Preview | 8.63 / 10 | 8.35 | $42.27 | 102x more expensive |
| GLM-5.3 | 8.36 / 10 | 8.12 | $48.25 | 117x more expensive |
| Qwen 3.8 Max | 8.20 / 10 | 7.85 | $48.55 | 117x more expensive |
| Qwen 3.7 Plus | 7.95 / 10 | 7.77 | $5.10 | 12x more expensive |
| Xiaomi MiMo V2.6 Flash | 7.98 / 10 | 7.69 | $0.81 | 2x more expensive |
| Gemini 3.8 Flash | 7.96 / 10 | 7.61 | $5.52 | 13x more expensive |
| Qwen 3.8 Flash | 7.72 / 10 | 7.30 | $9.44 | 23x more expensive |
| Claude Sonnet 5.5 | 7.83 / 10 | 7.51 | $9.00 | 22x more expensive |
| Gemini 3.5 Flash Lite | 5.92 / 10 | 5.61 | $0.89 | 2.1x more expensive |
| NVIDIA Nemotron-3 Nano 30B-A3B | 5.77 / 10 | 5.30 | $1.02 | 2.5x more expensive |
| NVIDIA Nemotron-3 Ultra 550B | 7.38 / 10 | 7.13 | $4.35 | 11x more expensive |
| NVIDIA Nemotron-3 Super 120B | 7.10 / 10 | 6.80 | $1.24 | 3x more expensive |
Cost breakdown
| Model | Quality | Confidence | Cost / 1k runs | Overpay | Mode |
|---|---|---|---|---|---|
| GPT-6 Luna ★ OpenAI | 8.54 / 10 CI [8.32, 8.75] | HIGH | $0.41 | best value | batch |
| Claude Haiku 5.5 Anthropic | 8.16 / 10 CI [7.87, 8.46] | HIGH | $0.50 | 1.2x | batch |
| MiniMax M3 OpenRouter | 8.38 / 10 CI [8.25, 8.51] | RANKED | $1.53 | 3.7x | batch |
| DeepSeek V4.1 Flash DeepSeek | 8.13 / 10 CI [7.81, 8.45] | MEDIUM | $1.57 | 3.8x | batch |
| Tencent Hy3 OpenRouter | 8.69 / 10 CI [8.52, 8.86] | RANKED | $3.44 | 8.3x | batch |
| Thinking Machines Inkling Small OpenRouter | 8.32 / 10 CI [8.01, 8.64] | MEDIUM | $4.50 | 11x | batch |
| GLM-5.3 Flash Z.AI | 8.46 / 10 CI [8.28, 8.64] | RANKED | $5.66 | 14x | batch |
| GPT-6 Sol OpenAI | 8.24 / 10 CI [7.96, 8.52] | HIGH | $6.67 | 16x | batch |
| Claude Opus 5.5 Anthropic | 8.14 / 10 CI [7.82, 8.46] | MEDIUM | $21.94 | 53x | batch |
| Grok 4.7 xAI | 8.52 / 10 CI [8.25, 8.79] | HIGH | $26.80 | 65x | batch |
| Meta Muse Spark 1.3 best OpenRouter | 8.97 / 10 CI [8.82, 9.13] | RANKED | $27.69 | 67x | batch |
| Thinking Machines Inkling OpenRouter | 8.29 / 10 CI [8.01, 8.57] | HIGH | $37.32 | 90x | batch |
| Moonshot Kimi K3 Moonshot AI | 8.68 / 10 CI [8.47, 8.90] | HIGH | $40.24 | 97x | batch |
| Tencent Hy4 Preview OpenRouter | 8.63 / 10 CI [8.35, 8.90] | HIGH | $42.27 | 102x | batch |
| GLM-5.3 Z.AI | 8.36 / 10 CI [8.12, 8.60] | HIGH | $48.25 | 117x | batch |
| Qwen 3.8 Max Alibaba Cloud (DashScope) | 8.20 / 10 CI [7.85, 8.54] | MEDIUM | $48.55 | 117x | 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: 4421 input tokens → 1305 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 correct isolation of primary content, removal of irrelevant page furniture and promotional noise, source fidelity, coverage of material facts and arguments, preservation of quantitative details, corrections, disclosures, warnings, and caveats, usefulness of organization, absence of invention, and appropriate compression. Penalize summaries that repeat navigation or advertisements, omit qualifying material as boilerplate, or reward brevity at the expense of decision-relevant information. Schema validity is deterministic.
Output schema
Every answer on this task is checked against this JSON Schema, whichever model wrote it. An answer that doesn't fit counts as a model failure, and the call is retried on another model.
{
"description": "Output schema for content summarization LLM call.\n\nThe summary should be comprehensive and capture ALL information\nrelevant to the capability's report structure. No arbitrary length limits.",
"properties": {
"summary": {
"description": "Comprehensive summary of the content. Extract ALL facts, data points, quotes, and insights relevant to the report structure requirements. No length limit - capture everything relevant.",
"title": "Summary",
"type": "string"
}
},
"required": [
"summary"
],
"title": "ContentSummaryResult",
"type": "object"
}Prompt templates
The published template for this task.
default ·
LLMB_STRUCTURED_CONTENT_SUMMARIZATION_SYSTEM +
LLMB_STRUCTURED_CONTENT_SUMMARIZATION_USER
(41843 calls in window)
System prompt
Identify and summarize the supplied content’s substantive material for downstream analysis. Treat layout and formatting as possible structural evidence, not automatically as content. Exclude page chrome, navigation, cookie or consent banners, subscription prompts, advertisements, promotional inserts, unrelated recommendations, social widgets, duplicated headers or footers, formatting artifacts, and other material unrelated to the primary content. Preserve bylines, dates, captions, footnotes, corrections, disclosures, warnings, disclaimers, and source context when they materially qualify or explain the substantive content. Prioritize faithful coverage of material facts, arguments, evidence, caveats, and conclusions over brevity. Use summary_structure to organize and prioritize content when supplied, but do not invent coverage for empty areas. Preserve important names, dates, numbers, units, uncertainty, and disagreements, and distinguish the source’s assertions from the summarizer’s organization. Treat empty optional values as absent and return only the requested result. Your response must conform exactly to this output schema: {schema_json_string}.
User prompt
Inputs — summary_structure: {summary_structure}; content_title: {content_title}; content_url: {content_url}; content_type: {content_type}; content_text: {content_text}. Use only these inputs to complete the task defined by the system prompt.