Best LLMs for Multi Perspective Decision Synthesis
Synthesizes multiple analyses and adversarial perspectives into a structured decision, confidence assessment, and optional constrained action plan. Illustrative uses include synthesizing evidence for vendor selection, product prioritization, software architecture, market entry, c
Run this task on Fronset — request an invitation
Models
Frontier on this task: Moonshot Kimi K3 at 8.84 / 10. Quality bar at 90%: 7.96.
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 |
|---|---|---|---|---|
| DeepSeek V4.1 Flash | 8.08 / 10 | 7.76 | $3.76 | best value |
| Xiaomi MiMo V2.6 Pro | 7.96 / 10 | 7.55 | $7.75 | 2.1x more expensive |
| Qwen 3.7 Plus | 8.28 / 10 | 8.12 | $12.92 | 3.4x more expensive |
| GLM-5.3 Flash | 8.55 / 10 | 8.33 | $25.24 | 6.7x more expensive |
| Thinking Machines Inkling Small | 8.46 / 10 | 8.22 | $38.30 | 10x more expensive |
| Tencent Hy4 Preview | 8.30 / 10 | 8.08 | $56.87 | 15x more expensive |
| Claude Sonnet 5.5 | 8.34 / 10 | 8.04 | $60.98 | 16x more expensive |
| Thinking Machines Inkling | 8.59 / 10 | 8.33 | $70.03 | 19x more expensive |
| Meta Muse Spark 1.3 | 8.13 / 10 | 7.94 | $74.35 | 20x more expensive |
| Grok 4.7 | 8.05 / 10 | 7.70 | $95.37 | 25x more expensive |
| Claude Opus 5.5 | 8.44 / 10 | 8.09 | $124.60 | 33x more expensive |
| Qwen 3.8 Max | 7.99 / 10 | 7.76 | $138.13 | 37x more expensive |
| Moonshot Kimi K3 | 8.84 / 10 | 8.70 | $156.13 | 41x more expensive |
| GLM-5.3 | 8.59 / 10 | 8.44 | $228.96 | 61x more expensive |
| GPT-6 Luna | 7.66 / 10 | 7.42 | $1.81 | 52% cheaper |
| Xiaomi MiMo V2.6 Flash | 7.78 / 10 | 7.41 | $2.46 | 35% cheaper |
| NVIDIA Nemotron-3 Super 120B | 7.21 / 10 | 6.97 | $6.19 | 1.6x more expensive |
| NVIDIA Nemotron-3 Nano 30B-A3B | 5.74 / 10 | 5.32 | $2.67 | 29% cheaper |
| Gemini 3.5 Flash Lite | 7.16 / 10 | 6.82 | $5.61 | 1.5x more expensive |
| Tencent Hy3 | 7.71 / 10 | 7.46 | $5.86 | 1.6x more expensive |
| NVIDIA Nemotron-3 Ultra 550B | 7.95 / 10 | 7.65 | $38.84 | 10x more expensive |
| GPT-6 Sol | 7.64 / 10 | 7.28 | $34.29 | 9.1x more expensive |
| NVIDIA Nemotron 3.5 Lightning | 6.58 / 10 | 6.14 | $3.75 | best value |
| MiniMax M3 | 7.72 / 10 | 7.58 | $14.62 | 3.9x more expensive |
| Gemini 3.8 Flash | 7.23 / 10 | 6.97 | $16.42 | 4.4x more expensive |
| Qwen 3.8 Flash | 6.71 / 10 | 6.34 | $20.68 | 5.5x more expensive |
Cost breakdown
| Model | Quality | Confidence | Cost / 1k runs | Overpay | Mode |
|---|---|---|---|---|---|
| DeepSeek V4.1 Flash ★ DeepSeek | 8.08 / 10 CI [7.76, 8.41] | MEDIUM | $3.76 | best value | batch |
| Xiaomi MiMo V2.6 Pro Xiaomi MiMo | 7.96 / 10 CI [7.55, 8.36] | MEDIUM | $7.75 | 2.1x | batch |
| Qwen 3.7 Plus Alibaba Cloud (DashScope) | 8.28 / 10 CI [8.12, 8.44] | RANKED | $12.92 | 3.4x | batch |
| GLM-5.3 Flash Z.AI | 8.55 / 10 CI [8.33, 8.77] | HIGH | $25.24 | 6.7x | batch |
| Thinking Machines Inkling Small OpenRouter | 8.46 / 10 CI [8.22, 8.71] | HIGH | $38.30 | 10x | batch |
| Tencent Hy4 Preview OpenRouter | 8.30 / 10 CI [8.08, 8.52] | HIGH | $56.87 | 15x | batch |
| Claude Sonnet 5.5 Anthropic | 8.34 / 10 CI [8.04, 8.64] | MEDIUM | $60.98 | 16x | batch |
| Thinking Machines Inkling OpenRouter | 8.59 / 10 CI [8.33, 8.85] | HIGH | $70.03 | 19x | batch |
| Meta Muse Spark 1.3 OpenRouter | 8.13 / 10 CI [7.94, 8.31] | RANKED | $74.35 | 20x | batch |
| Grok 4.7 xAI | 8.05 / 10 CI [7.70, 8.40] | MEDIUM | $95.37 | 25x | batch |
| Claude Opus 5.5 Anthropic | 8.44 / 10 CI [8.09, 8.80] | MEDIUM | $124.60 | 33x | batch |
| Qwen 3.8 Max Alibaba Cloud (DashScope) | 7.99 / 10 CI [7.76, 8.22] | HIGH | $138.13 | 37x | batch |
| Moonshot Kimi K3 best Moonshot AI | 8.84 / 10 CI [8.70, 8.98] | RANKED | $156.13 | 41x | batch |
| GLM-5.3 Z.AI | 8.59 / 10 CI [8.44, 8.73] | RANKED | $228.96 | 61x | 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: 42004 input tokens → 1429 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 faithful synthesis of all panel evidence, treatment of disagreement, direction and confidence calibration, constraint compliance, internal consistency, risk awareness, traceability, and absence of invented market facts.
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": "Individual voter's structured decision from the investment panel.",
"properties": {
"confidence": {
"description": "Confidence level in the vote (0.0 to 1.0)",
"maximum": 1,
"minimum": 0,
"title": "Confidence",
"type": "number"
},
"direction": {
"description": "Investment direction recommendation",
"enum": [
"BULLISH",
"BEARISH",
"NEUTRAL"
],
"title": "Direction",
"type": "string"
},
"entry_condition": {
"default": "",
"description": "Entry condition or price level (e.g., 'Buy at market', 'Limit order at $480', 'On pullback to 50-day MA')",
"title": "Entry Condition",
"type": "string"
},
"expected_pct_change": {
"description": "Expected percentage change (e.g., 12.5 for +12.5%, -8.0 for -8%)",
"title": "Expected Pct Change",
"type": "number"
},
"expected_timeframe_days": {
"description": "Expected timeframe in days for the move to materialize",
"minimum": 1,
"title": "Expected Timeframe Days",
"type": "integer"
},
"key_reasoning": {
"description": "2-3 sentence summary of the key reasoning behind this vote",
"title": "Key Reasoning",
"type": "string"
},
"position_size_pct": {
"default": 2,
"description": "Recommended position size as percentage of portfolio (e.g., 2.0 for 2%)",
"maximum": 100,
"minimum": 0,
"title": "Position Size Pct",
"type": "number"
},
"preferred_instrument": {
"default": "",
"description": "Specific tradeable instrument recommended (e.g., 'SPY', 'TSLA 200 call Apr 2025', '/ES futures')",
"title": "Preferred Instrument",
"type": "string"
},
"risk_factors": {
"description": "Top risk factors considered in reaching this decision",
"items": {
"type": "string"
},
"title": "Risk Factors",
"type": "array"
},
"stop_loss": {
"default": "",
"description": "Stop-loss level or condition (e.g., 'Stop at $460', 'Trail 5%', 'Exit if breaks below 200-day MA')",
"title": "Stop Loss",
"type": "string"
}
},
"required": [
"direction",
"expected_pct_change",
"expected_timeframe_days",
"confidence",
"key_reasoning"
],
"title": "InvestPanelVote",
"type": "object"
}Prompt templates
The published template for this task.
default ·
LLMB_MULTI_PERSPECTIVE_DECISION_SYNTHESIS_SYSTEM +
LLMB_MULTI_PERSPECTIVE_DECISION_SYNTHESIS_USER
(516 calls in window)
System prompt
Evaluate all supplied perspectives, evidence, disagreements, and challenges before deciding. Apply analysis_profile as the sole source of decision domain, labels, confidence scale, constraints, allowed actions, quantitative framework, risk treatment, and prohibited conclusions. Separate evidence from stakeholder preference, expose unresolved conflicts, and do not convert missing evidence into confidence. 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 — decision_context: {decision_context}; perspectives: {perspectives}; challenges: {challenges}; current_state: {current_state}; constraints: {constraints}; supporting_summary: {supporting_summary}; decision_notes: {decision_notes}; analysis_profile: {analysis_profile}. Use only these inputs to complete the task defined by the system prompt.