Best LLMs for Topic Cluster Labeling
Assigns a concise descriptive label and short explanation to a cluster of semantically related claims or content items. Illustrative uses include labeling clusters of customer requests, support issues, software defects, sales notes, research claims, interview excerpts, patient fe
Run this task on Fronset — request an invitation
Models
Frontier on this task: Moonshot Kimi K3 at 8.91 / 10. Quality bar at 90%: 8.01.
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 | 8.28 / 10 | 7.95 | $0.88 | best value |
| GPT-6 Luna | 8.32 / 10 | 8.05 | $0.88 | 1x more expensive |
| DeepSeek V4.1 Flash | 8.24 / 10 | 8.00 | $1.62 | 1.9x more expensive |
| Gemini 3.8 Flash | 8.15 / 10 | 7.92 | $1.80 | 2.1x more expensive |
| MiniMax M3 | 8.18 / 10 | 8.02 | $2.42 | 2.8x more expensive |
| Tencent Hy3 | 8.13 / 10 | 7.96 | $6.40 | 7.3x more expensive |
| Qwen 3.7 Plus | 8.13 / 10 | 7.79 | $6.55 | 7.5x more expensive |
| Thinking Machines Inkling Small | 8.29 / 10 | 7.92 | $7.74 | 8.8x more expensive |
| GPT-6 Sol | 8.50 / 10 | 8.29 | $9.30 | 11x more expensive |
| Claude Sonnet 5.5 | 8.42 / 10 | 8.18 | $16.03 | 18x more expensive |
| Meta Muse Spark 1.3 | 8.17 / 10 | 7.89 | $27.30 | 31x more expensive |
| Thinking Machines Inkling | 8.30 / 10 | 7.90 | $30.93 | 35x more expensive |
| Claude Opus 5.5 | 8.61 / 10 | 8.43 | $31.53 | 36x more expensive |
| Grok 4.7 | 8.27 / 10 | 7.98 | $38.04 | 43x more expensive |
| Moonshot Kimi K3 | 8.91 / 10 | 8.74 | $47.21 | 54x more expensive |
| GLM-5.3 | 8.65 / 10 | 8.41 | $87.48 | 100x more expensive |
| Claude Haiku 5.5 | 7.76 / 10 | 7.36 | $0.85 | 3% cheaper |
| Qwen 3.8 Flash | 7.64 / 10 | 7.24 | $7.92 | 9x more expensive |
| Qwen 3.8 Max | 7.90 / 10 | 7.53 | $51.51 | 59x more expensive |
| NVIDIA Nemotron-3 Nano 30B-A3B | 6.97 / 10 | 6.56 | $1.64 | 1.9x more expensive |
| Gemini 3.5 Flash Lite | 7.33 / 10 | 6.99 | $2.41 | 2.8x more expensive |
| Xiaomi MiMo V2.6 Pro | 7.91 / 10 | 7.42 | $2.11 | 2.4x more expensive |
Cost breakdown
| Model | Quality | Confidence | Cost / 1k runs | Overpay | Mode |
|---|---|---|---|---|---|
| Xiaomi MiMo V2.6 Flash ★ Xiaomi MiMo | 8.28 / 10 CI [7.95, 8.62] | MEDIUM | $0.88 | best value | batch |
| GPT-6 Luna OpenAI | 8.32 / 10 CI [8.05, 8.58] | HIGH | $0.88 | 1x | batch |
| DeepSeek V4.1 Flash DeepSeek | 8.24 / 10 CI [8.00, 8.48] | HIGH | $1.62 | 1.9x | batch |
| Gemini 3.8 Flash Gemini | 8.15 / 10 CI [7.92, 8.38] | HIGH | $1.80 | 2.1x | batch |
| MiniMax M3 OpenRouter | 8.18 / 10 CI [8.02, 8.33] | RANKED | $2.42 | 2.8x | batch |
| Tencent Hy3 OpenRouter | 8.13 / 10 CI [7.96, 8.30] | RANKED | $6.40 | 7.3x | batch |
| Qwen 3.7 Plus Alibaba Cloud (DashScope) | 8.13 / 10 CI [7.79, 8.48] | MEDIUM | $6.55 | 7.5x | batch |
| Thinking Machines Inkling Small OpenRouter | 8.29 / 10 CI [7.92, 8.66] | MEDIUM | $7.74 | 8.8x | batch |
| GPT-6 Sol OpenAI | 8.50 / 10 CI [8.29, 8.72] | HIGH | $9.30 | 11x | batch |
| Claude Sonnet 5.5 Anthropic | 8.42 / 10 CI [8.18, 8.67] | HIGH | $16.03 | 18x | batch |
| Meta Muse Spark 1.3 OpenRouter | 8.17 / 10 CI [7.89, 8.45] | HIGH | $27.30 | 31x | batch |
| Thinking Machines Inkling OpenRouter | 8.30 / 10 CI [7.90, 8.70] | MEDIUM | $30.93 | 35x | batch |
| Claude Opus 5.5 Anthropic | 8.61 / 10 CI [8.43, 8.80] | RANKED | $31.53 | 36x | batch |
| Grok 4.7 xAI | 8.27 / 10 CI [7.98, 8.57] | HIGH | $38.04 | 43x | batch |
| Moonshot Kimi K3 best Moonshot AI | 8.91 / 10 CI [8.74, 9.08] | RANKED | $47.21 | 54x | batch |
| GLM-5.3 Z.AI | 8.65 / 10 CI [8.41, 8.90] | HIGH | $87.48 | 100x | 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: 11880 input tokens → 366 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 representativeness, specificity, distinction from neighboring concepts, support across cluster members, label clarity, concise boundary description, and absence of unsupported interpretation.
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": "Output from the topic cluster naming LLM call.\n\nGiven a cluster of semantically similar claims and their synthesis summary,\nnames and describes the topic that the cluster represents.",
"properties": {
"topic_description": {
"default": "",
"description": "Brief description of what this topic covers and why these claims belong together (max 500 characters)",
"maxLength": 500,
"title": "Topic Description",
"type": "string"
},
"topic_name": {
"description": "Concise, descriptive name for the topic (e.g., 'Revenue Growth Outlook', 'Regulatory Headwinds')",
"title": "Topic Name",
"type": "string"
}
},
"required": [
"topic_name"
],
"title": "TopicClusterNamingOutput",
"type": "object"
}Prompt templates
The published template for this task.
default ·
LLMB_TOPIC_CLUSTER_LABELING_SYSTEM +
LLMB_TOPIC_CLUSTER_LABELING_USER
(4031 calls in window)
System prompt
Name the shared substantive idea that distinguishes this cluster from adjacent clusters. The name must be reusable across subjects: state the theme in wording that a cluster about a different company, market or programme could match into, and keep company, organisation, product, ticker and person names out of it — subject_name is context for reading the claims, never a component of the name. The specifics belong in the description: name there the entities, mechanisms, outcomes or tensions supported by multiple members, and state the common scope and the boundary that separates this cluster from its neighbours. Avoid generic labels such as “Overview,” unsupported conclusions, category repetition, and wording derived from a single outlier. 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 — subject_name: {subject_name}; category: {category}; claim_count: {claim_count}; claims_text: {claims_text}. Use only these inputs to complete the task defined by the system prompt.