Best LLMs for Profiled Document Analysis
Analyzes a supplied document using caller-provided document conventions, evidence rules, analytical dimensions, materiality, and decision context. Illustrative uses include analyzing contract risk, vendor documentation, software requirements, market disclosures, media-rights agre
Run this task on Fronset — request an invitation
Models
Frontier on this task: Grok 4.7 at 9.17 / 10. Score needed at the Balanced step: 8.26.
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 |
|---|---|---|---|---|
| Claude Haiku 5.5 | 8.33 / 10 | 7.84 | $1.03 | best value |
| DeepSeek V4.1 Flash | 8.45 / 10 | 7.98 | $1.89 | 1.8x more expensive |
| GPT-6 Sol | 8.58 / 10 | 8.30 | $6.82 | 6.6x more expensive |
| Meta Muse Spark 1.3 | 8.75 / 10 | 8.51 | $14.85 | 14x more expensive |
| Claude Sonnet 5.5 | 8.50 / 10 | 8.06 | $22.77 | 22x more expensive |
| Thinking Machines Inkling | 8.35 / 10 | 8.04 | $22.92 | 22x more expensive |
| Claude Opus 5.5 | 8.76 / 10 | 8.55 | $25.39 | 25x more expensive |
| Tencent Hy4 Preview | 8.91 / 10 | 8.65 | $43.30 | 42x more expensive |
| Grok 4.7 | 9.17 / 10 | 8.75 | $44.35 | 43x more expensive |
| Qwen 3.8 Max | 8.55 / 10 | 8.09 | $63.61 | 62x more expensive |
| Moonshot Kimi K3 | 8.79 / 10 | 8.37 | $102.74 | 99x more expensive |
| GLM-5.3 | 8.79 / 10 | 8.32 | $217.46 | 211x more expensive |
| Gemini 3.5 Flash Lite | 6.88 / 10 | 6.48 | $1.14 | 1.1x more expensive |
| NVIDIA Nemotron-3 Super 120B | 6.26 / 10 | 5.91 | $1.27 | 1.2x more expensive |
| Qwen 3.7 Plus | 7.25 / 10 | 6.76 | $21.86 | 21x more expensive |
| Gemini 3.8 Flash | 8.19 / 10 | 7.95 | $4.31 | 4.2x more expensive |
| MiniMax M3 | 8.04 / 10 | 7.63 | $7.08 | 6.9x more expensive |
| NVIDIA Nemotron-3 Ultra 550B | 7.15 / 10 | 6.72 | $5.63 | 5.5x more expensive |
| NVIDIA Nemotron-3 Nano 30B-A3B | 4.66 / 10 | 4.30 | $0.78 | 24% cheaper |
| Tencent Hy3 | 6.84 / 10 | 6.44 | $0.74 | 28% cheaper |
| Xiaomi MiMo V2.6 Pro | 7.51 / 10 | 7.03 | $1.28 | 1.2x more expensive |
Cost breakdown
| Model | Quality | Confidence | Cost / 1k runs | Overpay | Mode |
|---|---|---|---|---|---|
| Claude Haiku 5.5 ★ Anthropic | 8.33 / 10 CI [7.84, 8.82] | MEDIUM | $1.03 | best value | batch |
| DeepSeek V4.1 Flash DeepSeek | 8.45 / 10 CI [7.98, 8.92] | MEDIUM | $1.89 | 1.8x | batch |
| GPT-6 Sol OpenAI | 8.58 / 10 CI [8.30, 8.86] | HIGH | $6.82 | 6.6x | batch |
| Meta Muse Spark 1.3 OpenRouter | 8.75 / 10 CI [8.51, 8.99] | HIGH | $14.85 | 14x | batch |
| Claude Sonnet 5.5 Anthropic | 8.50 / 10 CI [8.06, 8.94] | MEDIUM | $22.77 | 22x | batch |
| Thinking Machines Inkling OpenRouter | 8.35 / 10 CI [8.04, 8.67] | MEDIUM | $22.92 | 22x | batch |
| Claude Opus 5.5 Anthropic | 8.76 / 10 CI [8.55, 8.97] | HIGH | $25.39 | 25x | batch |
| Tencent Hy4 Preview OpenRouter | 8.91 / 10 CI [8.65, 9.18] | HIGH | $43.30 | 42x | batch |
| Grok 4.7 best xAI | 9.17 / 10 CI [8.75, 9.59] | MEDIUM | $44.35 | 43x | batch |
| Qwen 3.8 Max Alibaba Cloud (DashScope) | 8.55 / 10 CI [8.09, 9.01] | MEDIUM | $63.61 | 62x | batch |
| Moonshot Kimi K3 Moonshot AI | 8.79 / 10 CI [8.37, 9.21] | MEDIUM | $102.74 | 99x | batch |
| GLM-5.3 Z.AI | 8.79 / 10 CI [8.32, 9.25] | MEDIUM | $217.46 | 211x | batch |
Overpay shows how much more you pay than the best-value model that meets the quality step (marked ★) — the best-value good-enough option. "16x" means you overpay 16× — 16× that reference for no quality benefit the step asks for. Typical call shape for this task: 2818 input tokens → 5112 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 $/1M list rates — this is what running the task costs, not a per-token price.
Evaluation rubric
Judge document-profile fidelity, factual and quantitative accuracy, materiality under the supplied analysis profile, preservation of qualifiers, separation of fact from interpretation, balance, source traceability, and usefulness to the stated decision context.
Output schema
Every answer on this task is checked against this JSON Schema, whichever model wrote it. An answer that doesn't fit counts as a model failure, and the call is retried on another model.
{
"$defs": {
"KeyEvent": {
"properties": {
"event_date": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"description": "The date of the event",
"title": "Event Date"
},
"event_summary": {
"description": "A concise summary of the key details",
"title": "Event Summary",
"type": "string"
},
"event_type": {
"description": "The category of the event",
"title": "Event Type",
"type": "string"
}
},
"required": [
"event_type",
"event_summary"
],
"title": "KeyEvent",
"type": "object"
}
},
"description": "Full SEC filing analysis result.",
"properties": {
"company_name": {
"title": "Company Name",
"type": "string"
},
"document_title": {
"title": "Document Title",
"type": "string"
},
"filing_date": {
"title": "Filing Date",
"type": "string"
},
"financial_highlights": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"title": "Financial Highlights"
},
"form_type": {
"title": "Form Type",
"type": "string"
},
"investment_implications": {
"title": "Investment Implications",
"type": "string"
},
"key_events_or_disclosures": {
"items": {
"$ref": "#/$defs/KeyEvent"
},
"title": "Key Events Or Disclosures",
"type": "array"
},
"management_discussion_summary": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"title": "Management Discussion Summary"
},
"overall_summary": {
"title": "Overall Summary",
"type": "string"
},
"period_of_report": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"title": "Period Of Report"
},
"risk_factor_summary": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"title": "Risk Factor Summary"
}
},
"required": [
"company_name",
"document_title",
"form_type",
"filing_date",
"overall_summary",
"key_events_or_disclosures",
"investment_implications"
],
"title": "SecFilingAnalysis",
"type": "object"
}Prompt templates
The published template for this task.
default ·
LLMB_PROFILED_DOCUMENT_ANALYSIS_SYSTEM +
LLMB_PROFILED_DOCUMENT_ANALYSIS_USER
(436 calls in window)
System prompt
Use only the supplied document and cited sources. Apply document_profile for terminology and legal or structural qualifiers and analysis_profile for the requested dimensions and decision lens. Distinguish source fact, source-party assertion, calculation, and reviewer interpretation; preserve dates, units, comparisons, status, and uncertainty. Identify missing context rather than filling it from memory. Derive document family, subject type, decision context, and permitted conclusions only from the supplied inputs and profiles. 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 — document_content: {document_content}; document_profile: {document_profile}; analysis_profile: {analysis_profile}. Use only these inputs to complete the task defined by the system prompt.