Best LLMs for Report Executive Summary Generation
Produces a concise executive summary of a long-form report or document, covering its most decision-relevant findings, trends, risks, opportunities, and recommendations within a caller-supplied length target. Illustrative uses include summarizing a board paper, due-diligence asses
Run this task on Fronset — request an invitation
Models
Frontier on this task: GLM-5.3 at 8.60 / 10. Quality bar at 90%: 7.74.
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 | 8.06 / 10 | 7.75 | $2.14 | best value |
| Xiaomi MiMo V2.6 Pro | 8.07 / 10 | 7.84 | $7.47 | 3.5x more expensive |
| Qwen 3.7 Plus | 8.18 / 10 | 7.98 | $15.32 | 7.1x more expensive |
| Gemini 3.8 Flash | 7.99 / 10 | 7.71 | $23.24 | 11x more expensive |
| GLM-5.3 Flash | 8.34 / 10 | 8.04 | $29.87 | 14x more expensive |
| GPT-6 Sol | 8.16 / 10 | 7.83 | $42.37 | 20x more expensive |
| Tencent Hy4 Preview | 7.87 / 10 | 7.41 | $68.82 | 32x more expensive |
| Claude Sonnet 5.5 | 8.06 / 10 | 7.77 | $76.00 | 35x more expensive |
| Grok 4.7 | 8.11 / 10 | 7.86 | $119.93 | 56x more expensive |
| Moonshot Kimi K3 | 7.97 / 10 | 7.74 | $136.38 | 64x more expensive |
| Claude Opus 5.5 | 8.23 / 10 | 7.99 | $150.80 | 70x more expensive |
| Qwen 3.8 Max | 8.35 / 10 | 8.07 | $198.18 | 92x more expensive |
| GLM-5.3 | 8.60 / 10 | 8.42 | $287.88 | 134x more expensive |
| Thinking Machines Inkling Small | 7.69 / 10 | 7.35 | $24.84 | 12x more expensive |
| Meta Muse Spark 1.3 | 7.72 / 10 | 7.34 | $74.41 | 35x more expensive |
| DeepSeek V4.1 Flash | 7.44 / 10 | 7.00 | $6.37 | 3x more expensive |
| Claude Haiku 5.5 | 7.70 / 10 | 7.46 | $4.09 | 1.9x more expensive |
| Gemini 3.5 Flash Lite | 6.89 / 10 | 6.42 | $7.49 | 3.5x more expensive |
| MiniMax M3 | 7.61 / 10 | 7.46 | $22.61 | 11x more expensive |
| Thinking Machines Inkling | 7.67 / 10 | 7.33 | $68.08 | 32x more expensive |
Cost breakdown
| Model | Quality | Confidence | Cost / 1k runs | Overpay | Mode |
|---|---|---|---|---|---|
| GPT-6 Luna ★ OpenAI | 8.06 / 10 CI [7.75, 8.36] | MEDIUM | $2.14 | best value | batch |
| Xiaomi MiMo V2.6 Pro Xiaomi MiMo | 8.07 / 10 CI [7.84, 8.30] | HIGH | $7.47 | 3.5x | batch |
| Qwen 3.7 Plus Alibaba Cloud (DashScope) | 8.18 / 10 CI [7.98, 8.37] | RANKED | $15.32 | 7.1x | batch |
| Gemini 3.8 Flash Gemini | 7.99 / 10 CI [7.71, 8.27] | HIGH | $23.24 | 11x | batch |
| GLM-5.3 Flash Z.AI | 8.34 / 10 CI [8.04, 8.65] | MEDIUM | $29.87 | 14x | batch |
| GPT-6 Sol OpenAI | 8.16 / 10 CI [7.83, 8.50] | MEDIUM | $42.37 | 20x | batch |
| Tencent Hy4 Preview OpenRouter | 7.87 / 10 CI [7.41, 8.33] | MEDIUM | $68.82 | 32x | batch |
| Claude Sonnet 5.5 Anthropic | 8.06 / 10 CI [7.77, 8.34] | HIGH | $76.00 | 35x | batch |
| Grok 4.7 xAI | 8.11 / 10 CI [7.86, 8.37] | HIGH | $119.93 | 56x | batch |
| Moonshot Kimi K3 Moonshot AI | 7.97 / 10 CI [7.74, 8.21] | HIGH | $136.38 | 64x | batch |
| Claude Opus 5.5 Anthropic | 8.23 / 10 CI [7.99, 8.46] | HIGH | $150.80 | 70x | batch |
| Qwen 3.8 Max Alibaba Cloud (DashScope) | 8.35 / 10 CI [8.07, 8.63] | HIGH | $198.18 | 92x | batch |
| GLM-5.3 best Z.AI | 8.60 / 10 CI [8.42, 8.78] | RANKED | $287.88 | 134x | 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: 49170 input tokens → 6153 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 coverage of decision-relevant content, factual fidelity, prioritization, preservation of caveats, usefulness to the stated audience, coherence, and compliance with the semantic length target. Exact word count 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": "Defines the structure for executive summary generation from consolidated chapters.",
"properties": {
"executive_summary": {
"description": "A concise executive summary (matching the target word count specified in the prompt) highlighting key insights, trends, risks, opportunities, and recommendations.",
"title": "Executive Summary",
"type": "string"
}
},
"required": [
"executive_summary"
],
"title": "ExecutiveSummary",
"type": "object"
}Prompt templates
The published template for this task.
default ·
LLMB_REPORT_EXECUTIVE_SUMMARY_GENERATION_SYSTEM +
LLMB_REPORT_EXECUTIVE_SUMMARY_GENERATION_USER
(19 calls in window)
System prompt
Summarize the complete supplied document for a time-constrained decision-maker. Prioritize material findings and their implications; preserve uncertainty, disagreement, and important quantitative details. Include risks, opportunities, and recommendations only when the source supports them. Do not introduce external facts or describe omitted sections as if they were analyzed. 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 — subject_name: {subject_name}; subject_code: {subject_code}; report_requirements: {report_requirements}; chapters_word_count: {chapters_word_count}; consolidated_chapters: {consolidated_chapters}; target_word_count: {target_word_count}; target_percentage: {target_percentage}; title_context: {title_context}; content: {content}. Use only these inputs to complete the task defined by the system prompt.