Cost mode:

Category: Evidence-Grounded Hypothesis Generation · Typical I/O: 2947→4147 tokens

Run this task on Fronset — request an invitation

Models

Frontier on this task: Moonshot Kimi K3 at 9.01 / 10. Quality bar at 90%: 8.11.

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
GPT-5.6 Luna8.62 / 108.31$0.84best value
DeepSeek V4 Flash9.01 / 108.78$3.854.6x more expensive
Thinking Machines Inkling Small8.62 / 108.24$5.726.8x more expensive
MiniMax M38.84 / 108.55$6.407.6x more expensive
Gemini 3.8 Flash8.58 / 108.26$7.549x more expensive
Qwen 3.7 Plus8.62 / 108.29$7.839.3x more expensive
GPT-5.6 Terra8.66 / 108.39$8.4910x more expensive
GPT-5.6 Sol8.75 / 108.48$12.7415x more expensive
Meta Muse Spark 1.38.86 / 108.61$16.0019x more expensive
Gemini 3.5 Flash8.45 / 108.05$19.7623x more expensive
Thinking Machines Inkling8.71 / 108.37$25.0130x more expensive
DeepSeek V4 Pro8.58 / 108.20$26.1431x more expensive
Grok 4.68.66 / 108.26$34.9842x more expensive
Tencent Hy4 Preview8.84 / 108.60$58.0169x more expensive
Moonshot Kimi K39.01 / 108.81$89.12106x more expensive
Qwen 3.8 Max8.18 / 107.69$122.59146x more expensive
Gemini 3.5 Flash Lite7.76 / 107.28$0.901.1x more expensive

Cost breakdown

ModelQualityConfidenceCost / 1k runsOverpayMode
GPT-5.6 Luna OpenAI8.62 / 10 CI [8.31, 8.92]MEDIUM$0.84best valuebatch
DeepSeek V4 Flash DeepSeek9.01 / 10 CI [8.78, 9.24]HIGH$3.854.6xbatch
Thinking Machines Inkling Small OpenRouter8.62 / 10 CI [8.24, 9.00]MEDIUM$5.726.8xbatch
MiniMax M3 OpenRouter8.84 / 10 CI [8.55, 9.12]HIGH$6.407.6xbatch
Gemini 3.8 Flash Gemini8.58 / 10 CI [8.26, 8.91]MEDIUM$7.549xbatch
Qwen 3.7 Plus Alibaba Cloud (DashScope)8.62 / 10 CI [8.29, 8.95]MEDIUM$7.839.3xbatch
GPT-5.6 Terra OpenAI8.66 / 10 CI [8.39, 8.93]HIGH$8.4910xbatch
GPT-5.6 Sol OpenAI8.75 / 10 CI [8.48, 9.01]HIGH$12.7415xbatch
Meta Muse Spark 1.3 OpenRouter8.86 / 10 CI [8.61, 9.10]HIGH$16.0019xbatch
Gemini 3.5 Flash Gemini8.45 / 10 CI [8.05, 8.84]MEDIUM$19.7623xbatch
Thinking Machines Inkling OpenRouter8.71 / 10 CI [8.37, 9.05]MEDIUM$25.0130xbatch
DeepSeek V4 Pro DeepSeek8.58 / 10 CI [8.20, 8.97]MEDIUM$26.1431xbatch
Grok 4.6 xAI8.66 / 10 CI [8.26, 9.05]MEDIUM$34.9842xbatch
Tencent Hy4 Preview OpenRouter8.84 / 10 CI [8.60, 9.09]HIGH$58.0169xbatch
Moonshot Kimi K3 best Moonshot AI9.01 / 10 CI [8.81, 9.22]HIGH$89.12106xbatch
Qwen 3.8 Max Alibaba Cloud (DashScope)8.18 / 10 CI [7.69, 8.66]MEDIUM$122.59146xbatch

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: 2947 input tokens → 4147 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.

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.

{
  "$defs": {
    "JournalHypothesisV1": {
      "additionalProperties": false,
      "description": "One thesis the article supports, and the claims behind it.",
      "properties": {
        "claim_refs": {
          "description": "1-based indices into the numbered evidence block, naming the claims this hypothesis rests on.",
          "items": {
            "type": "integer"
          },
          "title": "Claim Refs",
          "type": "array"
        },
        "headline": {
          "description": "The hypothesis in one line — a claim, not a topic.",
          "title": "Headline",
          "type": "string"
        },
        "rationale": {
          "description": "Why the cited claims support it, in two or three sentences.",
          "title": "Rationale",
          "type": "string"
        }
      },
      "required": [
        "headline",
        "rationale"
      ],
      "title": "JournalHypothesisV1",
      "type": "object"
    }
  },
  "additionalProperties": false,
  "description": "Every hypothesis drawn from one article.",
  "properties": {
    "hypotheses": {
      "description": "The requested number of distinct, non-overlapping hypotheses.",
      "items": {
        "$ref": "#/$defs/JournalHypothesisV1"
      },
      "title": "Hypotheses",
      "type": "array"
    }
  },
  "required": [
    "hypotheses"
  ],
  "title": "JournalHypothesesV1",
  "type": "object"
}

Prompt templates

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

LLMB_BENCH_ARTICLE_HYPOTHESIS_GENERATION_SYSTEM + LLMB_BENCH_ARTICLE_HYPOTHESIS_GENERATION_USER (1293 calls in window)

System prompt

Read one benchmark journal article together with the approved evidence it was written from, and state the small number of distinct hypotheses the article supports. A hypothesis is a claim someone could agree or disagree with, not a topic or a summary. Ground every hypothesis in the numbered evidence and cite the entries it rests on by their numbers; never assert a figure the evidence does not carry, and never restate a figure without the denominator it is given with. Prefer hypotheses that differ in kind rather than in wording, so that each one could be argued on its own. Say nothing about publication, channels or promotion. Your response must conform exactly to this output schema: {schema_json_string}.

User prompt

Produce exactly {hypothesis_count} hypotheses.

Article title: {title}

Article excerpt: {excerpt}

Article body:
{article_body}

Approved evidence (cite these by number):
{evidence_block}
JSON_REPAIR_SYSTEM + JSON_REPAIR_USER (12 calls in window)

System prompt

You are a JSON repair tool. The user gives you malformed or partial model output and a JSON Schema. Return ONLY a single valid JSON object that satisfies the schema, salvaging as much real content from the input as possible. Do not invent data for fields the input doesn't support — use the schema's allowed empty/null values. Output the JSON object only: no prose, no markdown, no code fences.

User prompt

JSON Schema:
{schema_json}

Malformed output to repair:
{raw_text}

Return only the corrected JSON object.