Cost mode:

Category: Content Summarization & Synthesis · Rail: absolute · Typical I/O: 3218→4500 tokens

Models

Frontier on this task: GLM-5.3 at 9.17 / 10. Quality bar at 90%: 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 clears the bar but whose CI low does not.

ModelQuality scoreCI lowCost / 1k runsvs best value
GLM-5.3 Flash9.04 / 108.92$4.77best value
MiniMax M38.87 / 108.78$5.821.2x more expensive
NVIDIA Nemotron-3 Ultra 550B8.27 / 107.98$6.351.3x more expensive
Gemini 3.8 Flash8.45 / 108.25$7.691.6x more expensive
Thinking Machines Inkling Small8.54 / 108.36$7.931.7x more expensive
Qwen 3.7 Plus8.59 / 108.43$8.811.8x more expensive
Gemini 3.5 Flash8.65 / 108.46$12.412.6x more expensive
Meta Muse Spark 1.38.63 / 108.49$14.793.1x more expensive
DeepSeek V4 Pro8.27 / 108.13$15.073.2x more expensive
Thinking Machines Inkling8.58 / 108.41$22.714.8x more expensive
Claude Opus 58.59 / 108.37$26.915.6x more expensive
Claude Sonnet 58.75 / 108.62$35.757.5x more expensive
Tencent Hy4 Preview8.91 / 108.77$36.767.7x more expensive
Grok 4.68.99 / 108.88$69.0314x more expensive
GLM-5.39.17 / 109.05$85.3018x more expensive
Moonshot Kimi K39.04 / 108.95$88.9619x more expensive
Claude Haiku 4.58.17 / 108.02$5.631.2x more expensive
DeepSeek V4 Flash8.22 / 108.06$6.371.3x more expensive
Gemini 3.1 Flash Lite7.92 / 107.77$1.3672% cheaper
Gemini 3.5 Flash Lite8.21 / 108.02$1.0877% cheaper
GPT-5.4 Nano8.04 / 107.81$1.8461% cheaper
NVIDIA Nemotron-3 Super 120B8.16 / 107.99$4.76best value
Qwen 3.8 Max8.19 / 107.78$78.8517x more expensive
NVIDIA Nemotron 3.5 Lightning7.45 / 107.00$2.6245% cheaper
NVIDIA Nemotron-3 Nano 30B-A3B7.78 / 107.57$1.1376% cheaper
Qwen 3.8 Flash7.77 / 107.40$4.408% cheaper
Tencent Hy38.20 / 107.90$2.2752% cheaper

Cost breakdown

ModelQualityConfidenceCost / 1k runsOverpayMode
GLM-5.3 Flash Z.AI9.04 / 10 CI [8.92, 9.16]RANKED$4.77best valuebatch
MiniMax M3 OpenRouter8.87 / 10 CI [8.78, 8.95]RANKED$5.821.2xbatch
NVIDIA Nemotron-3 Ultra 550B OpenRouter8.27 / 10 CI [7.98, 8.56]HIGH$6.351.3xbatch
Gemini 3.8 Flash Gemini8.45 / 10 CI [8.25, 8.66]HIGH$7.691.6xbatch
Thinking Machines Inkling Small OpenRouter8.54 / 10 CI [8.36, 8.71]RANKED$7.931.7xbatch
Qwen 3.7 Plus Alibaba Cloud (DashScope)8.59 / 10 CI [8.43, 8.74]RANKED$8.811.8xbatch
Gemini 3.5 Flash Gemini8.65 / 10 CI [8.46, 8.83]RANKED$12.412.6xbatch
Meta Muse Spark 1.3 OpenRouter8.63 / 10 CI [8.49, 8.77]RANKED$14.793.1xbatch
DeepSeek V4 Pro DeepSeek8.27 / 10 CI [8.13, 8.42]RANKED$15.073.2xbatch
Thinking Machines Inkling OpenRouter8.58 / 10 CI [8.41, 8.75]RANKED$22.714.8xbatch
Claude Opus 5 Anthropic8.59 / 10 CI [8.37, 8.81]HIGH$26.915.6xbatch
Claude Sonnet 5 Anthropic8.75 / 10 CI [8.62, 8.88]RANKED$35.757.5xbatch
Tencent Hy4 Preview OpenRouter8.91 / 10 CI [8.77, 9.04]RANKED$36.767.7xbatch
Grok 4.6 xAI8.99 / 10 CI [8.88, 9.09]RANKED$69.0314xbatch
GLM-5.3 best Z.AI9.17 / 10 CI [9.05, 9.30]RANKED$85.3018xbatch
Moonshot Kimi K3 Moonshot AI9.04 / 10 CI [8.95, 9.14]RANKED$88.9619xbatch

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: 3218 input tokens → 4500 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.

Prompt templates

This is a pooled capability — 2 prompt families share it. The pair shown first is the most frequently used in production.

PUBLICATION_TITLE_GENERATION_SYSTEM_PROMPT + PUBLICATION_TITLE_GENERATION_USER_PROMPT (485 calls in window)

System prompt

You are an expert Editor-in-Chief specializing in publication headlines.
Your task is to generate compelling, professional titles and subtitles for research publications.

{audience_context}

For each category provided, generate one title/subtitle pair that:
- Is accurate and reflects the content
- Is engaging and click-worthy without being clickbait
- Follows the category's specific angle/style
- Title: 5-12 words
- Subtitle: 10-20 words providing additional context

After generating all variants, act as the Editor-in-Chief:
- Select the single best variant as your "Editor's Choice"
- Explain in 2-3 sentences WHY this category and title best fits the article's core value and data

For the Editor's Choice, also generate two SEO fields:
- meta_title: The headline as it will appear in Google search results and the
  browser tab. Target ≤60 characters — beyond that, Google truncates and the
  meaning is lost. It must stand alone (the reader sees no subtitle, no image,
  no body). Front-load the subject and the outcome; cut filler words like
  "A Look At", "Insights On", "Exploring". The chosen `title` field can be
  longer and more expressive — `meta_title` is the search-snippet version.
- meta_description: The single-sentence summary shown under the title in
  Google search results. Target ≤155 characters. Present tense, summarises
  the takeaway, no clickbait, no trailing ellipsis.

## Required Output Format
Your response MUST be a single, valid JSON object conforming to this schema:
```json
{schema_json_string}
```

User prompt

Generate title/subtitle variants for the following analysis about "{title_context}":

CONTENT:
{content}

Generate one title/subtitle pair for each of the following categories:

{categories_text}

Then select the best variant as your Editor's Choice with a rationale explaining why it's the best fit.

The required JSON output schema is provided in the system prompt.
LLMB_PUBLICATION_TITLE_PACKAGE_GENERATION_SYSTEM + LLMB_PUBLICATION_TITLE_PACKAGE_GENERATION_USER (117 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.