Best LLMs for Promotional Message Generation
Generates one grounded promotional message for a content item using caller-supplied channel, audience, voice, length, link, and formatting rules. Illustrative uses include promoting a software release, product update, webinar, customer report, industry event, documentary, scienti
Run this task on Fronset — request an invitation
Models
Frontier on this task: Claude Opus 5.5 at 8.71 / 10. Quality bar at 90%: 7.84.
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 | 7.85 / 10 | 7.60 | $0.28 | best value |
| Claude Haiku 5.5 | 8.02 / 10 | 7.80 | $0.38 | 1.3x more expensive |
| Xiaomi MiMo V2.6 Flash | 7.87 / 10 | 7.49 | $0.45 | 1.6x more expensive |
| Xiaomi MiMo V2.6 Pro | 8.10 / 10 | 7.86 | $0.48 | 1.7x more expensive |
| DeepSeek V4.1 Flash | 7.86 / 10 | 7.50 | $0.83 | 3x more expensive |
| NVIDIA Nemotron-3 Super 120B | 7.88 / 10 | 7.64 | $1.14 | 4x more expensive |
| Gemini 3.8 Flash | 7.99 / 10 | 7.80 | $3.21 | 11x more expensive |
| Claude Sonnet 5.5 | 8.68 / 10 | 8.47 | $4.06 | 14x more expensive |
| Thinking Machines Inkling Small | 8.15 / 10 | 7.83 | $4.42 | 16x more expensive |
| GPT-6 Sol | 8.23 / 10 | 7.96 | $4.46 | 16x more expensive |
| GLM-5.3 Flash | 8.43 / 10 | 8.11 | $5.50 | 19x more expensive |
| Claude Opus 5.5 | 8.71 / 10 | 8.50 | $8.05 | 28x more expensive |
| Meta Muse Spark 1.3 | 8.03 / 10 | 7.68 | $11.62 | 41x more expensive |
| Thinking Machines Inkling | 8.42 / 10 | 8.26 | $16.73 | 59x more expensive |
| Moonshot Kimi K3 | 8.54 / 10 | 8.36 | $26.96 | 95x more expensive |
| Tencent Hy4 Preview | 8.29 / 10 | 8.04 | $38.29 | 136x more expensive |
| Grok 4.7 | 8.16 / 10 | 7.83 | $44.75 | 158x more expensive |
| GLM-5.3 | 8.52 / 10 | 8.31 | $105.23 | 372x more expensive |
| Gemini 3.5 Flash Lite | 7.40 / 10 | 7.11 | $0.47 | 1.7x more expensive |
| NVIDIA Nemotron-3 Nano 30B-A3B | 7.35 / 10 | 7.13 | $0.95 | 3.4x more expensive |
| NVIDIA Nemotron-3 Ultra 550B | 7.57 / 10 | 7.25 | $2.46 | 8.7x more expensive |
| Qwen 3.8 Flash | 7.57 / 10 | 7.15 | $3.13 | 11x more expensive |
| Qwen 3.8 Max | 7.83 / 10 | 7.52 | $33.56 | 119x more expensive |
| Tencent Hy3 | 7.78 / 10 | 7.53 | $0.96 | 3.4x more expensive |
| Qwen 3.7 Plus | 7.37 / 10 | 6.98 | $7.34 | 26x more expensive |
Cost breakdown
| Model | Quality | Confidence | Cost / 1k runs | Overpay | Mode |
|---|---|---|---|---|---|
| GPT-6 Luna ★ OpenAI | 7.85 / 10 CI [7.60, 8.11] | HIGH | $0.28 | best value | batch |
| Claude Haiku 5.5 Anthropic | 8.02 / 10 CI [7.80, 8.25] | HIGH | $0.38 | 1.3x | batch |
| Xiaomi MiMo V2.6 Flash Xiaomi MiMo | 7.87 / 10 CI [7.49, 8.25] | MEDIUM | $0.45 | 1.6x | batch |
| Xiaomi MiMo V2.6 Pro Xiaomi MiMo | 8.10 / 10 CI [7.86, 8.34] | HIGH | $0.48 | 1.7x | batch |
| DeepSeek V4.1 Flash DeepSeek | 7.86 / 10 CI [7.50, 8.23] | MEDIUM | $0.83 | 3x | batch |
| NVIDIA Nemotron-3 Super 120B OpenRouter | 7.88 / 10 CI [7.64, 8.12] | HIGH | $1.14 | 4x | batch |
| Gemini 3.8 Flash Gemini | 7.99 / 10 CI [7.80, 8.17] | RANKED | $3.21 | 11x | batch |
| Claude Sonnet 5.5 Anthropic | 8.68 / 10 CI [8.47, 8.89] | HIGH | $4.06 | 14x | batch |
| Thinking Machines Inkling Small OpenRouter | 8.15 / 10 CI [7.83, 8.47] | MEDIUM | $4.42 | 16x | batch |
| GPT-6 Sol OpenAI | 8.23 / 10 CI [7.96, 8.50] | HIGH | $4.46 | 16x | batch |
| GLM-5.3 Flash Z.AI | 8.43 / 10 CI [8.11, 8.74] | MEDIUM | $5.50 | 19x | batch |
| Claude Opus 5.5 best Anthropic | 8.71 / 10 CI [8.50, 8.92] | HIGH | $8.05 | 28x | batch |
| Meta Muse Spark 1.3 OpenRouter | 8.03 / 10 CI [7.68, 8.38] | MEDIUM | $11.62 | 41x | batch |
| Thinking Machines Inkling OpenRouter | 8.42 / 10 CI [8.26, 8.59] | RANKED | $16.73 | 59x | batch |
| Moonshot Kimi K3 Moonshot AI | 8.54 / 10 CI [8.36, 8.73] | RANKED | $26.96 | 95x | batch |
| Tencent Hy4 Preview OpenRouter | 8.29 / 10 CI [8.04, 8.53] | HIGH | $38.29 | 136x | batch |
| Grok 4.7 xAI | 8.16 / 10 CI [7.83, 8.48] | MEDIUM | $44.75 | 158x | batch |
| GLM-5.3 Z.AI | 8.52 / 10 CI [8.31, 8.73] | HIGH | $105.23 | 372x | 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: 4744 input tokens → 994 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 grounding, specificity, promotional effectiveness without hype, platform-native style, clarity, and preservation of the same supported message across variants. Deterministic platform limits are out of scope.
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.
{
"properties": {
"bluesky_post_text": {
"description": "Engaging Bluesky post under 260 characters. MUST include the literal placeholder \u003cghost_url\u003e where the link should appear. For model-specific LLM benchmark posts, include model/provider hashtags such as #ClaudeSonnet, #Sonnet5, or #Anthropic when relevant. For stock/equity posts, include the ticker as a hashtag, e.g. #AAPL.",
"title": "Bluesky Post Text",
"type": "string"
}
},
"required": [
"bluesky_post_text"
],
"title": "AutoBlueskyPostOutput",
"type": "object"
}Prompt templates
The published template for this task.
default ·
LLMB_PROMOTIONAL_MESSAGE_GENERATION_SYSTEM +
LLMB_PROMOTIONAL_MESSAGE_GENERATION_USER
(1220 calls in window)
System prompt
Communicate one specific, source-grounded reason to engage with the content. Apply channel_profile and optional voice_profile without changing the underlying claim. Use channel_profile as the sole source of channel-specific length, formatting, link, label, mention, and line-break rules. Avoid clickbait, invented urgency, unsupported superlatives, engagement bait, and generic announcement filler. 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 — title: {title}; content: {content}; channel_profile: {channel_profile}; voice_profile: {voice_profile}. Use only these inputs to complete the task defined by the system prompt.