Best LLMs for Research Query Validation
Validates and minimally repairs a research query for a specified search platform and region without changing its intended information need. Illustrative uses include repairing queries for enterprise search, code search, e-discovery, market-intelligence platforms, newsroom researc
Run this task on Fronset — request an invitation
Models
Frontier on this task: Meta Muse Spark 1.3 at 8.68 / 10. Quality bar at 90%: 7.82.
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 | 8.42 / 10 | 7.93 | $0.08 | best value |
| Claude Haiku 5.5 | 8.43 / 10 | 7.96 | $0.22 | 2.7x more expensive |
| MiniMax M3 | 8.61 / 10 | 8.42 | $0.46 | 5.7x more expensive |
| Gemini 3.8 Flash | 8.67 / 10 | 8.39 | $1.74 | 22x more expensive |
| GLM-5.3 Flash | 8.54 / 10 | 8.15 | $3.27 | 41x more expensive |
| Meta Muse Spark 1.3 | 8.68 / 10 | 8.39 | $7.20 | 89x more expensive |
| Qwen 3.8 Max | 8.54 / 10 | 8.18 | $18.68 | 232x more expensive |
| GLM-5.3 | 8.28 / 10 | 7.80 | $42.21 | 523x more expensive |
| Xiaomi MiMo V2.6 Pro | 7.58 / 10 | 7.08 | $0.26 | 3.2x more expensive |
| Thinking Machines Inkling Small | 7.67 / 10 | 7.21 | $2.92 | 36x more expensive |
| NVIDIA Nemotron-3 Super 120B | 7.61 / 10 | 7.14 | $0.45 | 5.5x more expensive |
Cost breakdown
| Model | Quality | Confidence | Cost / 1k runs | Overpay | Mode |
|---|---|---|---|---|---|
| Xiaomi MiMo V2.6 Flash ★ Xiaomi MiMo | 8.42 / 10 CI [7.93, 8.92] | MEDIUM | $0.08 | best value | batch |
| Claude Haiku 5.5 Anthropic | 8.43 / 10 CI [7.96, 8.90] | MEDIUM | $0.22 | 2.7x | batch |
| MiniMax M3 OpenRouter | 8.61 / 10 CI [8.42, 8.80] | RANKED | $0.46 | 5.7x | batch |
| Gemini 3.8 Flash Gemini | 8.67 / 10 CI [8.39, 8.95] | HIGH | $1.74 | 22x | batch |
| GLM-5.3 Flash Z.AI | 8.54 / 10 CI [8.15, 8.92] | MEDIUM | $3.27 | 41x | batch |
| Meta Muse Spark 1.3 best OpenRouter | 8.68 / 10 CI [8.39, 8.98] | HIGH | $7.20 | 89x | batch |
| Qwen 3.8 Max Alibaba Cloud (DashScope) | 8.54 / 10 CI [8.18, 8.89] | MEDIUM | $18.68 | 232x | batch |
| GLM-5.3 Z.AI | 8.28 / 10 CI [7.80, 8.76] | MEDIUM | $42.21 | 523x | 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: 1309 input tokens → 878 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 syntax validity, preservation of search intent, minimality of repair, platform-rule correctness, retention of useful constraints, and explanation accuracy. Penalize needless rewrites and semantic broadening.
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 model for query validation and optimization",
"properties": {
"changes_made": {
"description": "List of changes applied to the original query",
"items": {
"type": "string"
},
"title": "Changes Made",
"type": "array"
},
"optimized_query": {
"description": "The validated and optimized query text (primary/first query if split)",
"title": "Optimized Query",
"type": "string"
},
"split_queries": {
"description": "If query was too complex, list of simpler split queries. Empty if no split needed.",
"items": {
"type": "string"
},
"title": "Split Queries",
"type": "array"
},
"validation_status": {
"default": "",
"description": "Status: valid, modified, split, or invalid",
"title": "Validation Status",
"type": "string"
},
"warnings": {
"description": "Any warnings about the query",
"items": {
"type": "string"
},
"title": "Warnings",
"type": "array"
}
},
"required": [
"optimized_query",
"changes_made"
],
"title": "QueryValidationOutput",
"type": "object"
}Prompt templates
The published template for this task.
default ·
LLMB_RESEARCH_QUERY_VALIDATION_SYSTEM +
LLMB_RESEARCH_QUERY_VALIDATION_USER
(0 calls in window)
System prompt
Check syntax and semantics against the target platform’s supplied rules.
Target platform rules:
{platform_rules}
Remove or translate unsupported operators, malformed grouping, invalid field filters, and accidental date restrictions while preserving the query’s intent and useful terms. Do not optimize by adding new subject matter. Return the original when it is already valid. 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 — query_text: {query_text}; search_platform: {search_platform}; target_region: {target_region}. Use only these inputs to complete the task defined by the system prompt.