What the steps mean
Cost mode:

Category: Topic Organization & Clustering · Typical I/O: 3353→6859 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: GLM-5.3 at 8.86 / 10. Score needed at the Balanced step: 7.98.

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.

ModelQuality scoreCI lowCost / 1k runsvs best value
Xiaomi MiMo V2.6 Flash8.08 / 107.83$0.24best value
Tencent Hy38.12 / 107.85$0.753.1x more expensive
Xiaomi MiMo V2.6 Pro8.42 / 108.16$1.104.5x more expensive
NVIDIA Nemotron-3 Ultra 550B8.09 / 107.72$4.4118x more expensive
Gemini 3.8 Flash7.98 / 107.68$6.0025x more expensive
GLM-5.3 Flash8.58 / 108.38$15.0462x more expensive
Meta Muse Spark 1.38.06 / 107.85$15.5164x more expensive
Grok 4.78.20 / 107.89$55.83230x more expensive
Moonshot Kimi K38.37 / 108.04$80.13331x more expensive
Qwen 3.8 Max8.07 / 107.77$102.42423x more expensive
GLM-5.38.86 / 108.65$173.25715x more expensive
Claude Opus 5.57.72 / 107.38$24.21100x more expensive
Tencent Hy4 Preview7.73 / 107.33$47.91198x more expensive
GPT-6 Sol7.17 / 106.88$7.5731x more expensive
Claude Haiku 5.57.11 / 106.77$0.743.1x more expensive
Qwen 3.8 Flash7.62 / 107.20$6.2726x more expensive
Gemini 3.5 Flash Lite6.95 / 106.67$1.797.4x more expensive
NVIDIA Nemotron-3 Nano 30B-A3B6.71 / 106.24$3.5415x more expensive
MiniMax M37.79 / 107.37$0.953.9x more expensive
NVIDIA Nemotron-3 Super 120B7.30 / 106.91$2.068.5x more expensive
Qwen 3.7 Plus7.80 / 107.40$8.9737x more expensive
Thinking Machines Inkling Small7.67 / 107.39$6.5927x more expensive
Thinking Machines Inkling7.88 / 107.59$29.04120x more expensive
DeepSeek V4.1 Flash7.86 / 107.58$3.7716x more expensive

Cost breakdown

ModelQualityConfidenceCost / 1k runsOverpayMode
Xiaomi MiMo V2.6 Flash ★ Xiaomi MiMo8.08 / 10 CI [7.83, 8.34]HIGH$0.24best valuebatch
Tencent Hy3 OpenRouter8.12 / 10 CI [7.85, 8.40]HIGH$0.753.1xbatch
Xiaomi MiMo V2.6 Pro Xiaomi MiMo8.42 / 10 CI [8.16, 8.68]HIGH$1.104.5xbatch
NVIDIA Nemotron-3 Ultra 550B OpenRouter8.09 / 10 CI [7.72, 8.45]MEDIUM$4.4118xbatch
Gemini 3.8 Flash Gemini7.98 / 10 CI [7.68, 8.27]HIGH$6.0025xbatch
GLM-5.3 Flash Z.AI8.58 / 10 CI [8.38, 8.77]RANKED$15.0462xbatch
Meta Muse Spark 1.3 OpenRouter8.06 / 10 CI [7.85, 8.26]HIGH$15.5164xbatch
Grok 4.7 xAI8.20 / 10 CI [7.89, 8.50]MEDIUM$55.83230xbatch
Moonshot Kimi K3 Moonshot AI8.37 / 10 CI [8.04, 8.69]MEDIUM$80.13331xbatch
Qwen 3.8 Max Alibaba Cloud (DashScope)8.07 / 10 CI [7.77, 8.37]MEDIUM$102.42423xbatch
GLM-5.3 best Z.AI8.86 / 10 CI [8.65, 9.07]HIGH$173.25715xbatch

Overpay shows how much more you pay than the best-value model that meets the quality step (marked ★) — the best-value good-enough option. "16x" means you overpay 16× — 16× that reference for no quality benefit the step asks for. Typical call shape for this task: 3353 input tokens → 6859 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 dependency ordering, narrative coherence, progression, local transitions, reader orientation, and use of every topic exactly once. Identifier completeness is 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.

{
  "$defs": {
    "TopicSequenceItem": {
      "description": "Single topic in the recommended sequence.",
      "properties": {
        "topic_id": {
          "description": "The ID of the topic (exactly as provided in input)",
          "title": "Topic Id",
          "type": "integer"
        },
        "topic_name": {
          "description": "The name of the topic (exactly as provided in input, do not modify)",
          "title": "Topic Name",
          "type": "string"
        }
      },
      "required": [
        "topic_id",
        "topic_name"
      ],
      "title": "TopicSequenceItem",
      "type": "object"
    }
  },
  "description": "Output model for determining the optimal sequence of topics in a report.\nThe LLM analyzes topic content and determines the best reading order.",
  "properties": {
    "sequence_reasoning": {
      "default": "",
      "description": "Brief explanation of why this sequence was chosen (e.g., logical flow, building complexity, narrative arc).",
      "title": "Sequence Reasoning",
      "type": "string"
    },
    "sequenced_topics": {
      "description": "Ordered list of topics in the recommended reading sequence. First topic appears first in report.",
      "items": {
        "$ref": "#/$defs/TopicSequenceItem"
      },
      "title": "Sequenced Topics",
      "type": "array"
    }
  },
  "required": [
    "sequenced_topics"
  ],
  "title": "TopicSequenceOutput",
  "type": "object"
}

Prompt templates

The published template for this task.

default · LLMB_TOPIC_SEQUENCE_OPTIMIZATION_SYSTEM + LLMB_TOPIC_SEQUENCE_OPTIMIZATION_USER (0 calls in window)

System prompt

Return a permutation of the supplied topics only. Place necessary foundations before dependent analysis, group related material without collapsing distinct topics, and build a deliberate opening, progression, and conclusion suited to the subject. Use topic descriptions as evidence; do not rewrite, add, drop, or duplicate topics. 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}; topic_count: {topic_count}; topics_json: {topics_json}. Use only these inputs to complete the task defined by the system prompt.