
Anthropic’s Claude Opus 5.5 accomplished a analysis task in beneath three hours that sometimes takes human contractors 16 hours, in keeping with Epoch AI.
The AI analysis group reported on Nov. 17 that the mannequin produced work similar to that of its skilled contractors, elevating the potential for automating components of its analysis operations.
The task concerned researching EcoDataCenter 1 for Epoch’s knowledge heart database, a course of requiring researchers to look at information stories, regulatory filings, and satellite tv for pc imagery to ascertain particulars in regards to the facility.
Epoch offered the AI fashions with its contractor pointers, analysis instruments, and analysis standards earlier than assessing their work towards the requirements utilized to human researchers.
Claude Opus 5.5 scored 87.5% on the task, finishing it in slightly below three hours at a decrease price than hiring a contractor. One other Anthropic mannequin, Claude Fable 5.2, delivered a fair stronger consequence, scoring 97.5%. Nonetheless, it required roughly six and a half hours and value greater than a human contractor.
Each fashions met Epoch’s requirements for core analysis, though their submissions nonetheless required revisions to tools classifications, picture annotations, and written descriptions.
Epoch’s analyst judged each first passes similar to contractor work and recognized revisions for every.
The 16-hour determine is typical contractor labor per website. Epoch leaves evaluate time, time to publication and run-specific imagery waits unspecified. Its price comparability provides the rating with out itemizing bills.
By comparability, OpenAI’s GPT-6.2 Astra scored 65% on the identical task, lacking vital info and estimates that Epoch stated the obtainable proof may have supported.
AI may change how analysis contractors work
The findings have prompted Epoch to rethink the way it allocates work between AI fashions and human contractors.
The group stated it’s contemplating shifting some contractor time away from gathering and compiling analysis towards reviewing AI-generated submissions and performing different knowledge heart evaluation.
Such a transition may let AI techniques deal with a lot of the preliminary analysis whereas human staff give attention to checking findings, correcting errors, and making ready the fabric for publication.
Nonetheless, the outcomes additionally uncovered variations within the economics of utilizing AI for skilled work.
Whereas Opus accomplished the task quicker and cheaper than a contractor, Fable’s higher-quality output price extra, regardless of additionally ending significantly quicker than a human researcher.
The fashions additionally struggled with components of the task that required visible interpretation, significantly labeling satellite tv for pc photos and figuring out tools.
These shortcomings assist clarify why Epoch has not proposed eliminating human involvement completely, at the same time as its fashions strategy contractor-level efficiency.
As an alternative, the group is contemplating a workflow during which AI produces the preliminary analysis and contractors refine the outcomes, probably decreasing the human labor required for every database entry.