The Aggregate — LLM benchmark aggregate
Fused LLM rankings: one IRT/Elo scale across ~5,000 public benchmark leaderboards, updated daily.
我該用這個嗎
品質與安全性
根據工具定義與協定合規性的自動化分析。
上下文成本
這是每次將伺服器的工具載入模型上下文時所消耗的約略 token 數量。數量越高，可用於其他工作的注意力就越少。
安裝
一鍵安裝
將以下內容加入你的 `claude_desktop_config.json` 檔案：
{
"mcpServers": {
"the-aggregate": {
"url": "https://theaggregate.ai/mcp"
}
}
}遠端端點
https://theaggregate.ai/mcpstreamable-http它能做什麼
工具清單
工具（8）
🟢get_leaderboard(limit, offset, include_variants)
Top of the cross-benchmark aggregate ranking: every model placed on one Elo scale by an IRT model fit over public benchmark leaderboards (call about_the_aggregate for the current coverage counts). One row per model by default, fused across reasoning-effort settings. Supports paging via limit/offset.
輸入結構描述
{
"type": "object",
"properties": {
"limit": {
"type": "number",
"description": "Rows to return (1-100, default 25)."
},
"offset": {
"type": "number",
"description": "Rows to skip from the top (default 0)."
},
"include_variants": {
"type": "boolean",
"description": "Rank each reasoning-effort variant separately (e.g. \"Claude Opus 4.6 (High)\") instead of one fused row per model. Default false."
}
}
}🟢search_models(query, limit, include_variants)
Find ranked models by (partial) name or provider. Returns rank, Elo and the model page URL. One row per model by default, fused across reasoning-effort settings.
輸入結構描述
{
"type": "object",
"properties": {
"query": {
"type": "string",
"description": "Model or provider name fragment, e.g. \"opus\" or \"deepseek\"."
},
"limit": {
"type": "number",
"description": "Max results (1-25, default 10)."
},
"include_variants": {
"type": "boolean",
"description": "Return each reasoning-effort variant separately (e.g. \"Claude Opus 4.6 (High)\") instead of one fused row per model. Default false."
}
},
"required": [
"query"
]
}🟢get_model(model)
One model in depth: aggregate rank, Elo with standard error, provider, what it is, cost per task where known, and its most notable benchmark results (with percentiles).
輸入結構描述
{
"type": "object",
"properties": {
"model": {
"type": "string",
"description": "Model name or slug, e.g. \"Claude Opus 4.5\" or \"gpt-5-5\"."
}
},
"required": [
"model"
]
}⚪compare_models(models)
Head-to-head between 2-4 models: aggregate ranks, Elo gap with a significance note based on the standard errors, and notable benchmarks they share.
輸入結構描述
{
"type": "object",
"properties": {
"models": {
"type": "array",
"items": {
"type": "string"
},
"minItems": 2,
"maxItems": 4,
"description": "Two to four model names or slugs."
}
},
"required": [
"models"
]
}🟢search_benchmarks(query, limit)
Find benchmarks in the aggregate by (partial) name. Returns model coverage, difficulty on the Elo scale, and the benchmark page URL.
輸入結構描述
{
"type": "object",
"properties": {
"query": {
"type": "string",
"description": "Benchmark name fragment, e.g. \"swe-bench\" or \"arena\"."
},
"limit": {
"type": "number",
"description": "Max results (1-25, default 10)."
}
},
"required": [
"query"
]
}🟢get_benchmark(benchmark, top)
One benchmark in depth: what it measures, the original source leaderboard URL, IRT stats (difficulty, noise, model coverage), skill weights, and the current top models on it.
輸入結構描述
{
"type": "object",
"properties": {
"benchmark": {
"type": "string",
"description": "Benchmark name or slug, e.g. \"Aider polyglot\"."
},
"top": {
"type": "number",
"description": "How many top models to list (1-50, default 10)."
}
},
"required": [
"benchmark"
]
}🟢get_prediction_duel
Guesswork — the public prediction duel: every day frontier LLMs and The Aggregate's own IRT model predict newly scraped benchmark scores before seeing them, and the errors are scored. Returns the current monthly standings, wins and losses included.
輸入結構描述
{
"type": "object",
"properties": {}
}🟡about_the_aggregate
What this data is: how the IRT fusion works, current coverage counts, update cadence, and how to cite it.
輸入結構描述
{
"type": "object",
"properties": {}
}社群
證據