Best LLMs for Social Post Portfolio Selection
Selects the best fixed-size portfolio of candidate social posts using quality, expected audience value, topical diversity, timeliness, and caller-supplied constraints. Illustrative uses include selecting a balanced portfolio of customer updates, product messages, developer-commun
Run this task on Fronset — request an invitation
Models
Frontier on this task: Gemini 3.8 Flash at 8.72 / 10. Quality bar at 90%: 7.85.
point-estimate floor (CI low) · upper CI (less certain) · Bars sorted by blended cost; best-value model first.
| Model | Quality score | CI low | Cost / 1k runs | vs best value |
|---|---|---|---|---|
| Tencent Hy3 | 7.98 / 10 | 7.77 | $0.20 | best value |
| Claude Haiku 5.5 | 8.64 / 10 | 8.44 | $0.21 | 1.1x more expensive |
| GPT-6 Luna | 8.55 / 10 | 8.29 | $0.27 | 1.4x more expensive |
| Gemini 3.5 Flash Lite | 8.56 / 10 | 8.32 | $0.30 | 1.5x more expensive |
| NVIDIA Nemotron-3 Nano 30B-A3B | 8.31 / 10 | 7.91 | $0.38 | 1.9x more expensive |
| Xiaomi MiMo V2.6 Pro | 8.41 / 10 | 8.20 | $0.55 | 2.8x more expensive |
| MiniMax M3 | 8.39 / 10 | 8.32 | $0.63 | 3.2x more expensive |
| Xiaomi MiMo V2.6 Flash | 8.66 / 10 | 8.49 | $0.90 | 4.6x more expensive |
| NVIDIA Nemotron 3.5 Lightning | 8.03 / 10 | 7.73 | $1.15 | 5.8x more expensive |
| NVIDIA Nemotron-3 Super 120B | 8.07 / 10 | 7.62 | $1.32 | 6.7x more expensive |
| NVIDIA Nemotron-3 Ultra 550B | 8.29 / 10 | 7.96 | $1.63 | 8.3x more expensive |
| GPT-6 Sol | 8.69 / 10 | 8.49 | $2.84 | 14x more expensive |
| Claude Sonnet 5.5 | 8.58 / 10 | 8.35 | $2.89 | 15x more expensive |
| DeepSeek V4.1 Flash | 8.47 / 10 | 8.30 | $3.04 | 15x more expensive |
| Gemini 3.8 Flash | 8.72 / 10 | 8.54 | $3.52 | 18x more expensive |
| Thinking Machines Inkling Small | 8.69 / 10 | 8.48 | $3.79 | 19x more expensive |
| Qwen 3.8 Flash | 8.22 / 10 | 8.01 | $4.04 | 20x more expensive |
| GLM-5.3 Flash | 8.66 / 10 | 8.53 | $4.10 | 21x more expensive |
| Qwen 3.7 Plus | 8.34 / 10 | 8.19 | $5.85 | 30x more expensive |
| Claude Opus 5.5 | 8.59 / 10 | 8.38 | $6.58 | 33x more expensive |
| Meta Muse Spark 1.3 | 8.69 / 10 | 8.53 | $12.13 | 62x more expensive |
| Grok 4.7 | 8.48 / 10 | 8.26 | $20.96 | 106x more expensive |
| Thinking Machines Inkling | 8.52 / 10 | 8.26 | $23.46 | 119x more expensive |
| Qwen 3.8 Max | 8.15 / 10 | 8.00 | $26.49 | 134x more expensive |
| GLM-5.3 | 8.59 / 10 | 8.42 | $31.58 | 160x more expensive |
| Moonshot Kimi K3 | 8.62 / 10 | 8.38 | $32.02 | 162x more expensive |
| Tencent Hy4 Preview | 8.64 / 10 | 8.48 | $33.11 | 168x more expensive |
Cost breakdown
| Model | Quality | Confidence | Cost / 1k runs | Overpay | Mode |
|---|---|---|---|---|---|
| Tencent Hy3 ★ OpenRouter | 7.98 / 10 CI [7.77, 8.18] | HIGH | $0.20 | best value | batch |
| Claude Haiku 5.5 Anthropic | 8.64 / 10 CI [8.44, 8.83] | RANKED | $0.21 | 1.1x | batch |
| GPT-6 Luna OpenAI | 8.55 / 10 CI [8.29, 8.81] | HIGH | $0.27 | 1.4x | batch |
| Gemini 3.5 Flash Lite Gemini | 8.56 / 10 CI [8.32, 8.81] | HIGH | $0.30 | 1.5x | batch |
| NVIDIA Nemotron-3 Nano 30B-A3B OpenRouter | 8.31 / 10 CI [7.91, 8.71] | MEDIUM | $0.38 | 1.9x | batch |
| Xiaomi MiMo V2.6 Pro Xiaomi MiMo | 8.41 / 10 CI [8.20, 8.62] | HIGH | $0.55 | 2.8x | batch |
| MiniMax M3 OpenRouter | 8.39 / 10 CI [8.32, 8.47] | RANKED | $0.63 | 3.2x | batch |
| Xiaomi MiMo V2.6 Flash Xiaomi MiMo | 8.66 / 10 CI [8.49, 8.82] | RANKED | $0.90 | 4.6x | batch |
| NVIDIA Nemotron 3.5 Lightning OpenRouter | 8.03 / 10 CI [7.73, 8.33] | MEDIUM | $1.15 | 5.8x | batch |
| NVIDIA Nemotron-3 Super 120B OpenRouter | 8.07 / 10 CI [7.62, 8.53] | MEDIUM | $1.32 | 6.7x | batch |
| NVIDIA Nemotron-3 Ultra 550B OpenRouter | 8.29 / 10 CI [7.96, 8.62] | MEDIUM | $1.63 | 8.3x | batch |
| GPT-6 Sol OpenAI | 8.69 / 10 CI [8.49, 8.89] | RANKED | $2.84 | 14x | batch |
| Claude Sonnet 5.5 Anthropic | 8.58 / 10 CI [8.35, 8.82] | HIGH | $2.89 | 15x | batch |
| DeepSeek V4.1 Flash DeepSeek | 8.47 / 10 CI [8.30, 8.64] | RANKED | $3.04 | 15x | batch |
| Gemini 3.8 Flash best Gemini | 8.72 / 10 CI [8.54, 8.90] | RANKED | $3.52 | 18x | batch |
| Thinking Machines Inkling Small OpenRouter | 8.69 / 10 CI [8.48, 8.91] | HIGH | $3.79 | 19x | batch |
| Qwen 3.8 Flash Alibaba Cloud (DashScope) | 8.22 / 10 CI [8.01, 8.43] | HIGH | $4.04 | 20x | batch |
| GLM-5.3 Flash Z.AI | 8.66 / 10 CI [8.53, 8.79] | RANKED | $4.10 | 21x | batch |
| Qwen 3.7 Plus Alibaba Cloud (DashScope) | 8.34 / 10 CI [8.19, 8.49] | RANKED | $5.85 | 30x | batch |
| Claude Opus 5.5 Anthropic | 8.59 / 10 CI [8.38, 8.81] | HIGH | $6.58 | 33x | batch |
| Meta Muse Spark 1.3 OpenRouter | 8.69 / 10 CI [8.53, 8.86] | RANKED | $12.13 | 62x | batch |
| Grok 4.7 xAI | 8.48 / 10 CI [8.26, 8.70] | HIGH | $20.96 | 106x | batch |
| Thinking Machines Inkling OpenRouter | 8.52 / 10 CI [8.26, 8.79] | HIGH | $23.46 | 119x | batch |
| Qwen 3.8 Max Alibaba Cloud (DashScope) | 8.15 / 10 CI [8.00, 8.30] | RANKED | $26.49 | 134x | batch |
| GLM-5.3 Z.AI | 8.59 / 10 CI [8.42, 8.76] | RANKED | $31.58 | 160x | batch |
| Moonshot Kimi K3 Moonshot AI | 8.62 / 10 CI [8.38, 8.85] | HIGH | $32.02 | 162x | batch |
| Tencent Hy4 Preview OpenRouter | 8.64 / 10 CI [8.48, 8.79] | RANKED | $33.11 | 168x | 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: 1924 input tokens → 546 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 selected-item quality, portfolio diversity, timeliness, relevance, avoidance of redundancy, identifier fidelity, and consistency with the requested count and constraints.
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": "LLM output for X post selection within daily budget.",
"properties": {
"selected_pspp_ids": {
"description": "IDs of the selected PSPPs to publish",
"items": {
"type": "string"
},
"title": "Selected Pspp Ids",
"type": "array"
}
},
"required": [
"selected_pspp_ids"
],
"title": "PostSelectionOutput",
"type": "object"
}Prompt templates
The published template for this task.
default ·
LLMB_SOCIAL_POST_PORTFOLIO_SELECTION_SYSTEM +
LLMB_SOCIAL_POST_PORTFOLIO_SELECTION_USER
(0 calls in window)
System prompt
Evaluate candidates individually, then select the strongest portfolio rather than merely the top independent scores. Balance quality and timeliness with topic and framing diversity; avoid near-duplicates and preserve candidate identifiers exactly. Do not infer engagement from author identity or unsupported metrics, and return exactly the requested number when enough eligible candidates exist. 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 — select_count: {select_count}; total_count: {total_count}; candidates_text: {candidates_text}. Use only these inputs to complete the task defined by the system prompt.