SPB Git

spb/llmindex Public

The discriminative, contamination-resistant, fully transparent LLM ranking — updated live.

TypeScript 77.9% TeX 15.2% Python 3.7% SQL 1.4% JavaScript 1.1% Shell 0.5%
2.7 KB · 70 lines tsx
Raw Blame History
1/**2 * llmindex.io — dedicated full-page live efficiency frontier3 * Author:  Simon-Pierre Boucher4 * Contact: contact@spboucher.ai5 * License: Proprietary — © Simon-Pierre Boucher, all rights reserved6 */7import type { Metadata } from 'next';8import Link from 'next/link';9import { FrontierChart, type FrontierPoint } from '@/components/FrontierChart';10import { getEfficiencyData } from '@/lib/data';1112export const metadata: Metadata = {13  title: 'Efficiency Frontier',14  description:15    'Live Pareto frontier of LLM quality vs. measured cost — every point with 95% confidence intervals, updated automatically as models finish the benchmark.',16};17export const revalidate = 300;1819export default async function FrontierPage() {20  const data = await getEfficiencyData();21  const initial: FrontierPoint[] = data.map((p) => ({22    model: p.slug,23    name: p.name,24    provider: p.provider,25    score: p.score,26    score_low: p.scoreLow,27    score_high: p.scoreHigh,28    latency_p50: p.latencyP50,29    cost_per_1k_items: Number(p.costPer1kItems.toFixed(4)),30    on_frontier: p.onFrontier,31  }));3233  return (34    <div className="space-y-6">35      <div className="space-y-2">36        <h1 className="text-2xl font-bold tracking-tight text-zinc-900 sm:text-3xl">37          Efficiency frontier38        </h1>39        <p className="max-w-3xl text-sm leading-relaxed text-zinc-600">40          Global Index versus <em>measured</em> cost per 1,000 evaluation items (metered tokens ×41          live pricing), on a log scale. Models on the{' '}42          <span className="font-semibold text-emerald-700">frontier</span> are not dominated on43          both axes; every other model is strictly worse on quality <em>and</em> cost than some44          frontier point. By design this is published as a set — never collapsed into a single45          blended “value” number. The chart updates itself as the live benchmark lands each model.46        </p>47      </div>4849      {/* full-bleed chart: break out of the content column on large screens */}50      <div className="lg:relative lg:left-1/2 lg:w-[min(96vw,1500px)] lg:-translate-x-1/2">51        <FrontierChart initial={initial} />52      </div>5354      <p className="text-sm text-zinc-500">55        Prefer numbers? The same data lives in the{' '}56        <Link href="/" className="underline hover:text-emerald-600">57          leaderboard table58        </Link>{' '}59        and the{' '}60        <a href="/api/v1/efficiency" className="underline hover:text-emerald-600">61          public API62        </a>63        . Methodology: cost is the mean measured cost per item across a model&apos;s scored64        domains × 1,000; whiskers are the 95% CI of the Global Index; the frontier is computed on65        (score ↑, cost ↓) dominance.66      </p>67    </div>68  );69}70