Loading prices…
🔥BULLISH

Claude Opus 5.5 Completes 16-Hour Research Job in 3 Hours

Epoch AI is considering a shift from outsourced research to human review, though the cost and quality of AI-generated work varied across models.

Claude Opus 5.5 Completes 16-Hour Research Job in 3 Hours
Claude Opus 5.5 Completes 16-Hour Research Job in 3 Hours

Anthropic's Claude Opus 5.5 completed a data center research assignment in just under three hours that typically takes human contractors 16 hours, Epoch AI said. The model scored 87.5% against Epoch's evaluation criteria and cost less than a human contractor. Its work was comparable to that of professional contractors, though it still needed revisions.

Why it matters

The assignment covered EcoDataCenter 1, a facility in Epoch's data center database. Researchers had to examine news reports, regulatory filings and satellite imagery to establish details about it. Epoch gave the models its contractor guidelines, research tools and evaluation criteria before applying the same standards used for human researchers.

Claude Fable 5.2 scored higher at 97.5%, but took about six and a half hours and cost more than a contractor. Both Anthropic models met Epoch's core research standards, yet needed corrections to equipment classifications, image annotations and written descriptions. OpenAI's GPT-6.2 Astra scored 65%, missing information and estimates Epoch said the evidence could have supported.

Market impact

The results highlight both the potential productivity gains and the uneven economics of automating professional research. Opus delivered comparable work faster and at lower cost, while Fable's higher score came at a higher cost than a human contractor.

Epoch is considering shifting some contractor time from gathering and compiling information to reviewing AI-generated submissions and other data center analysis. Human checks remain important, particularly for interpreting satellite images and identifying equipment.

Frequently asked questions

  1. What did Epoch AI ask Claude to research?

    The assignment involved researching EcoDataCenter 1 for Epoch's data center database using news reports, regulatory filings and satellite imagery.

  2. How did Claude Opus 5.5 compare with a human contractor?

    Opus scored 87.5% and completed the assignment in just under three hours at a lower cost. Epoch said the same task typically takes human contractors 16 hours.

  3. How did Claude Fable 5.2 perform on the assignment?

    Fable scored 97.5%, higher than Opus, but took about six and a half hours and cost more than a human contractor.

  4. What errors still required human review?

    The Anthropic models' submissions needed revisions to equipment classifications, image annotations and written descriptions. Visual interpretation was a particular challenge.

  5. How might Epoch change its use of research contractors?

    Epoch is considering shifting some contractor time away from gathering and compiling research toward reviewing AI-generated submissions and other data center analysis.

Source attribution
Aggregated from CryptoSlate · Verified · Last refreshed 7m ago
Open original →