Best LLMs for Publication Title Package Generation
Generates title and subtitle candidates for caller-defined editorial categories, selects a best candidate, and produces associated SEO title metadata from supplied content and audience context. Illustrative uses include creating titles for a white paper, customer case study, deve
Run this task on Fronset — request an invitation
Models
Frontier on this task: GLM-5.3 at 9.12 / 10. Quality bar at 90%: 8.21.
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 |
|---|---|---|---|---|
| Xiaomi MiMo V2.6 Flash | 8.41 / 10 | 8.13 | $0.36 | best value |
| GPT-6 Luna | 8.61 / 10 | 8.30 | $0.49 | 1.4x more expensive |
| Xiaomi MiMo V2.6 Pro | 8.40 / 10 | 8.06 | $0.88 | 2.4x more expensive |
| DeepSeek V4.1 Flash | 8.68 / 10 | 8.47 | $2.42 | 6.7x more expensive |
| MiniMax M3 | 8.79 / 10 | 8.70 | $4.25 | 12x more expensive |
| Qwen 3.7 Plus | 8.46 / 10 | 8.29 | $6.04 | 17x more expensive |
| Thinking Machines Inkling Small | 8.49 / 10 | 8.32 | $6.68 | 18x more expensive |
| Gemini 3.8 Flash | 8.24 / 10 | 8.05 | $7.62 | 21x more expensive |
| GPT-6 Sol | 8.98 / 10 | 8.74 | $8.17 | 23x more expensive |
| GLM-5.3 Flash | 9.00 / 10 | 8.89 | $11.19 | 31x more expensive |
| Claude Opus 5.5 | 8.73 / 10 | 8.49 | $15.92 | 44x more expensive |
| Meta Muse Spark 1.3 | 8.63 / 10 | 8.49 | $16.64 | 46x more expensive |
| Thinking Machines Inkling | 8.52 / 10 | 8.35 | $30.73 | 85x more expensive |
| Tencent Hy4 Preview | 8.74 / 10 | 8.59 | $38.38 | 106x more expensive |
| Grok 4.7 | 9.02 / 10 | 8.82 | $59.89 | 165x more expensive |
| Moonshot Kimi K3 | 9.00 / 10 | 8.91 | $65.39 | 181x more expensive |
| GLM-5.3 | 9.12 / 10 | 9.01 | $199.91 | 552x more expensive |
| Gemini 3.5 Flash Lite | 8.10 / 10 | 7.90 | $1.07 | 3x more expensive |
| NVIDIA Nemotron-3 Ultra 550B | 8.08 / 10 | 7.78 | $7.27 | 20x more expensive |
| NVIDIA Nemotron-3 Super 120B | 8.10 / 10 | 7.92 | $2.17 | 6x more expensive |
| Qwen 3.8 Max | 7.96 / 10 | 7.57 | $73.57 | 203x more expensive |
| NVIDIA Nemotron 3.5 Lightning | 7.36 / 10 | 6.93 | $2.00 | 5.5x more expensive |
| Qwen 3.8 Flash | 7.52 / 10 | 7.22 | $10.44 | 29x more expensive |
| NVIDIA Nemotron-3 Nano 30B-A3B | 7.66 / 10 | 7.43 | $1.13 | 3.1x more expensive |
| Tencent Hy3 | 8.13 / 10 | 7.84 | $2.10 | 5.8x more expensive |
| Claude Haiku 5.5 | 7.62 / 10 | 7.25 | $0.45 | 1.2x more expensive |
| Claude Sonnet 5.5 | 8.00 / 10 | 7.72 | $8.03 | 22x more expensive |
Cost breakdown
| Model | Quality | Confidence | Cost / 1k runs | Overpay | Mode |
|---|---|---|---|---|---|
| Xiaomi MiMo V2.6 Flash ★ Xiaomi MiMo | 8.41 / 10 CI [8.13, 8.68] | HIGH | $0.36 | best value | batch |
| GPT-6 Luna OpenAI | 8.61 / 10 CI [8.30, 8.92] | MEDIUM | $0.49 | 1.4x | batch |
| Xiaomi MiMo V2.6 Pro Xiaomi MiMo | 8.40 / 10 CI [8.06, 8.73] | MEDIUM | $0.88 | 2.4x | batch |
| DeepSeek V4.1 Flash DeepSeek | 8.68 / 10 CI [8.47, 8.89] | HIGH | $2.42 | 6.7x | batch |
| MiniMax M3 OpenRouter | 8.79 / 10 CI [8.70, 8.87] | RANKED | $4.25 | 12x | batch |
| Qwen 3.7 Plus Alibaba Cloud (DashScope) | 8.46 / 10 CI [8.29, 8.64] | RANKED | $6.04 | 17x | batch |
| Thinking Machines Inkling Small OpenRouter | 8.49 / 10 CI [8.32, 8.66] | RANKED | $6.68 | 18x | batch |
| Gemini 3.8 Flash Gemini | 8.24 / 10 CI [8.05, 8.44] | RANKED | $7.62 | 21x | batch |
| GPT-6 Sol OpenAI | 8.98 / 10 CI [8.74, 9.21] | HIGH | $8.17 | 23x | batch |
| GLM-5.3 Flash Z.AI | 9.00 / 10 CI [8.89, 9.11] | RANKED | $11.19 | 31x | batch |
| Claude Opus 5.5 Anthropic | 8.73 / 10 CI [8.49, 8.97] | HIGH | $15.92 | 44x | batch |
| Meta Muse Spark 1.3 OpenRouter | 8.63 / 10 CI [8.49, 8.77] | RANKED | $16.64 | 46x | batch |
| Thinking Machines Inkling OpenRouter | 8.52 / 10 CI [8.35, 8.70] | RANKED | $30.73 | 85x | batch |
| Tencent Hy4 Preview OpenRouter | 8.74 / 10 CI [8.59, 8.88] | RANKED | $38.38 | 106x | batch |
| Grok 4.7 xAI | 9.02 / 10 CI [8.82, 9.22] | RANKED | $59.89 | 165x | batch |
| Moonshot Kimi K3 Moonshot AI | 9.00 / 10 CI [8.91, 9.09] | RANKED | $65.39 | 181x | batch |
| GLM-5.3 best Z.AI | 9.12 / 10 CI [9.01, 9.23] | RANKED | $199.91 | 552x | 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: 3623 input tokens → 1270 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 $/1M list rates — this is what running the task costs, not a per-token price.
Evaluation rubric
Judge factual grounding, title specificity, category differentiation, audience fit, subtitle complementarity, editor-choice rationale, SEO faithfulness, and resistance to clickbait. Exact length limits are deterministic.
Output schema
Every answer on this task is checked against this JSON Schema, whichever model wrote it. An answer that doesn't fit counts as a model failure, and the call is retried on another model.
{
"$defs": {
"EditorChoice": {
"description": "The LLM's recommended best title from the generated variants.",
"properties": {
"meta_description": {
"description": "SEO meta description shown as the Google search snippet. Target ≤155 characters. One sentence, present tense, summarising the takeaway. No clickbait, no ellipses.",
"title": "Meta Description",
"type": "string"
},
"meta_title": {
"description": "SEO/social title used for the \u003ctitle\u003e tag and og:title. Target ≤60 characters — Google truncates beyond that in search results. Must convey the main idea standalone (assume the reader sees nothing else). Front-load subject + outcome; avoid filler ('A Look At...', 'Insights On...').",
"title": "Meta Title",
"type": "string"
},
"rationale": {
"description": "2-3 sentence explanation of why this title was chosen as the best fit for the content",
"title": "Rationale",
"type": "string"
},
"selected_category_code": {
"description": "The category_code of the recommended best title",
"title": "Selected Category Code",
"type": "string"
}
},
"required": [
"selected_category_code",
"rationale",
"meta_title",
"meta_description"
],
"title": "EditorChoice",
"type": "object"
},
"TitleVariant": {
"description": "A single title/subtitle variant for a specific category.",
"properties": {
"category_code": {
"description": "The category code this variant belongs to",
"title": "Category Code",
"type": "string"
},
"subtitle": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"description": "Publication subtitle (10-20 words)",
"title": "Subtitle"
},
"title": {
"description": "Publication title (5-12 words)",
"title": "Title",
"type": "string"
}
},
"required": [
"category_code",
"title"
],
"title": "TitleVariant",
"type": "object"
}
},
"description": "Output model for publication title generation.\nGenerates one title/subtitle per configured category,\nthen selects the best (Editor's Choice) with rationale.",
"properties": {
"editors_choice": {
"$ref": "#/$defs/EditorChoice",
"description": "The recommended best title variant with rationale"
},
"variants": {
"description": "One title/subtitle variant per configured category",
"items": {
"$ref": "#/$defs/TitleVariant"
},
"title": "Variants",
"type": "array"
}
},
"required": [
"variants",
"editors_choice"
],
"title": "TitleGenerationOutput",
"type": "object"
}Prompt templates
The published template for this task.
default ·
LLMB_PUBLICATION_TITLE_PACKAGE_GENERATION_SYSTEM +
LLMB_PUBLICATION_TITLE_PACKAGE_GENERATION_USER
(237 calls in window)
System prompt
Generate one genuinely distinct candidate per supplied category, each grounded in the content’s central value. Preserve important entities and avoid claims stronger than the source. Select the editor’s choice using audience fit, accuracy, specificity, and differentiation—not sensationalism. Produce SEO metadata that preserves the selected title’s meaning and respects caller limits. 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 — audience_context: {audience_context}; title_context: {title_context}; content: {content}; categories_text: {categories_text}. Use only these inputs to complete the task defined by the system prompt.