Best LLMs for Research Community Selection

Category: Relevance, Classification & Matching · Typical I/O: 7646→2498 tokens

Runs on model group: All models. Inputs may reach any model in the group, not only the ranked models below.

Run this task on Fronset — request an invitation

Models

Frontier on this task: Meta Muse Spark 1.3 at 7.78 / 10. Quality bar at 90%: 7.00.

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
DeepSeek V4.1 Flash	7.49 / 10	7.22	$2.45	best value
Qwen 3.7 Plus	7.60 / 10	7.25	$3.93	1.6x more expensive
Gemini 3.8 Flash	7.04 / 10	6.63	$5.06	2.1x more expensive
GPT-6 Sol	7.61 / 10	7.32	$6.47	2.6x more expensive
GLM-5.3 Flash	7.58 / 10	7.31	$8.10	3.3x more expensive
Claude Sonnet 5.5	7.09 / 10	6.73	$14.44	5.9x more expensive
Meta Muse Spark 1.3	7.78 / 10	7.51	$16.04	6.5x more expensive
Claude Opus 5.5	7.51 / 10	7.23	$16.13	6.6x more expensive
Grok 4.7	7.62 / 10	7.39	$36.61	15x more expensive
Qwen 3.8 Max	7.18 / 10	6.88	$58.46	24x more expensive
MiniMax M3	6.93 / 10	6.75	$1.08	56% cheaper
Moonshot Kimi K3	6.39 / 10	5.94	$65.10	27x more expensive
Xiaomi MiMo V2.6 Pro	6.81 / 10	6.55	$1.19	51% cheaper
Claude Haiku 5.5	6.37 / 10	5.92	$0.44	82% cheaper
Xiaomi MiMo V2.6 Flash	6.93 / 10	6.60	$0.31	87% cheaper

Cost breakdown

Model	Quality	Confidence	Cost / 1k runs	Overpay	Mode
DeepSeek V4.1 Flash ★ DeepSeek	7.49 / 10 CI [7.22, 7.75]	HIGH	$2.45	best value	batch
Qwen 3.7 Plus Alibaba Cloud (DashScope)	7.60 / 10 CI [7.25, 7.96]	MEDIUM	$3.93	1.6x	batch
Gemini 3.8 Flash Gemini	7.04 / 10 CI [6.63, 7.45]	MEDIUM	$5.06	2.1x	batch
GPT-6 Sol OpenAI	7.61 / 10 CI [7.32, 7.91]	HIGH	$6.47	2.6x	batch
GLM-5.3 Flash Z.AI	7.58 / 10 CI [7.31, 7.86]	HIGH	$8.10	3.3x	batch
Claude Sonnet 5.5 Anthropic	7.09 / 10 CI [6.73, 7.45]	MEDIUM	$14.44	5.9x	batch
Meta Muse Spark 1.3 best OpenRouter	7.78 / 10 CI [7.51, 8.04]	HIGH	$16.04	6.5x	batch
Claude Opus 5.5 Anthropic	7.51 / 10 CI [7.23, 7.79]	HIGH	$16.13	6.6x	batch
Grok 4.7 xAI	7.62 / 10 CI [7.39, 7.84]	HIGH	$36.61	15x	batch
Qwen 3.8 Max Alibaba Cloud (DashScope)	7.18 / 10 CI [6.88, 7.48]	MEDIUM	$58.46	24x	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: 7646 input tokens → 2498 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 subject and purpose relevance, community existence, expected research value, coverage/diversity, accessibility, appropriate specificity, and absence of fabricated activity or policy claims.

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.

{
  "$defs": {
    "Subreddit": {
      "description": "Individual subreddit",
      "properties": {
        "category": {
          "default": "",
          "description": "Subreddit category",
          "title": "Category",
          "type": "string"
        },
        "name": {
          "description": "Subreddit name (without r/ prefix)",
          "title": "Name",
          "type": "string"
        }
      },
      "required": [
        "name"
      ],
      "title": "Subreddit",
      "type": "object"
    }
  },
  "description": "Output model for subreddit selection",
  "properties": {
    "subreddits": {
      "description": "List of selected subreddits",
      "items": {
        "$ref": "#/$defs/Subreddit"
      },
      "title": "Subreddits",
      "type": "array"
    }
  },
  "required": [
    "subreddits"
  ],
  "title": "SubredditSelectionOutput",
  "type": "object"
}

Prompt templates

The published template for this task.

default · LLMB_RESEARCH_COMMUNITY_SELECTION_SYSTEM + LLMB_RESEARCH_COMMUNITY_SELECTION_USER (0 calls in window)

System prompt

Recommend communities likely to contain useful first-hand discussion or domain expertise for the stated research purpose. Apply channel_profile as the sole source of community types, identifier formats, accessibility states, and activity signals. Balance broad coverage with focused relevance and do not invent community names, activity, access, or policy facts. 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}; subject_type: {subject_type}; subject_description: {subject_description}; research_purpose: {research_purpose}; chapter_context: {chapter_context}; regions: {regions}; additional_requirements: {additional_requirements}; channel_profile: {channel_profile}. Use only these inputs to complete the task defined by the system prompt.