hanzo.ai/app/(marketing)/benchmarks/gpqa/page.tsx

/**
 * /benchmarks/gpqa — the router on GPQA-Diamond, against the field.
 *
 * The board is <Ladder />, read when the page opens from the research API's
 * `enso.gpqa_diamond` runs. Around it is what a reader arriving from a search
 * for "GPQA" needs: what the benchmark is, whose it is, what routing does to
 * it, and what a benchmark score does not establish. This file types no score.
 */
import { T } from '@hanzo/i18n/react'
import { pageMeta } from '@/lib/page-meta'
import { BENCHMARKS, bench } from '@/lib/benchmarks'
import { BenchJsonLd } from '@/components/benchmarks/JsonLd'
import { H2, Meta, More, P, Shell, Spec } from '@/components/benchmarks/kit'
import { Ladder } from '@/components/benchmarks/Ladder'

const B = bench('gpqa')!

export const metadata = pageMeta({
  title: 'GPQA-Diamond routing benchmark',
  description:
    'One set of 198 graduate-level questions through Enso\'s router and single models from other labs, every row a research run with its accuracy, count and what its tokens cost at list prices.',
  path: '/benchmarks/gpqa',
})

const MEASURES = ['GPQA-Diamond accuracy', 'what a run\'s tokens cost at list prices']

export default function Page() {
  return (
    <>
      <BenchJsonLd b={B} measures={MEASURES} />
      <Shell b={B}>
        <Spec b={B} />

        <H2><T>What is measured</T></H2>
        <P>
          <T>{'GPQA-Diamond is the hardest subset of GPQA: 198 multiple-choice questions in biology, physics and chemistry, written by domain PhDs and filtered to the ones that other PhDs get right and skilled non-experts get wrong even with the web open. It is a knowledge benchmark rather than an agentic one — one question, one answer, no tools — which is what makes it useful here: it isolates the choice of model from everything else a system does.'}</T>
        </P>
        <P>
          <T>{'What the router adds is that the choice is per question. Enso reads each one and dispatches it, so a tier is not a model — it is a pool and a willingness to spend. Every row below answers the same 198 questions, which is why a score and what its run cost belong in one table, and why a score quoted without its cost says nothing about the system it came from.'}</T>
        </P>

        <Ladder />

        <H2><T>What this does not establish</T></H2>
        <P>
          <strong>
            <T>A benchmark score is not a capability.</T>
          </strong>{' '}
          <T>
            {'GPQA-Diamond is four-way multiple choice on settled science. It measures what a pool knows and rewards a router that recognises a hard question; it says nothing about whether the answer is reasoned, whether the model holds up over a long autonomous run, or whether it can use a tool. The routing overhead is real work this table does not score either.'}
          </T>
        </P>
        <P>
          <strong>
            <T>A row is a run, as of its date.</T>
          </strong>{' '}
          <T>
            {"Every row went through enso-bench's harness at the tag and on the date it shows; a system is the router or a single model as that run met it, not necessarily a preset the catalog serves today. What a row cost is the tokens its run records at today's list price for its own model id, the router's included, so two rows are priced by one rule."}
          </T>
        </P>

        <More b={B} all={BENCHMARKS} />
        <Meta>
          <T
            values={{
              api: <a href="https://api.hanzo.ai/v1/research/runs?org=hanzo&amp;benchmark=enso.gpqa_diamond">api.hanzo.ai</a>,
              dataset: <a href={B.dataset!.url}>{B.dataset!.name}</a>,
              license: B.dataset!.license,
            }}
          >
            {'every figure a research run at {api} · the dataset: {dataset} under {license}'}
          </T>
        </Meta>
      </Shell>
    </>
  )
}