Epoch AI Benchmark Finds Six Leading LLMs Fall Short of Its Own Employees' Work Quality
Key Takeaways
- •Epoch AI's Automation Reports, released on October 8, 2026, found that none of the six tested large language models reached the 100% quality standard defined by the work of its own human employees.
- •Claude Fable 5.1 and GPT-6 Astra shared the top score at approximately 65% of human-level work quality, while Gemini 3.8 Flash trailed at 42%, leaving a gap of more than 20 percentage points.
- •The evaluation used open-ended tasks drawn from Epoch's real internal work, including graphic design and research design, with human graders scoring outputs against the organization's internal standards.
- •All tested models performed better on structured tasks such as coding and data analysis, but every model struggled with open-ended work requiring judgment or tailoring to a specific audience.
- •Model rankings showed no change between October 8 and November 17, 2026, and no single developer has decisively pulled ahead on this type of open-ended work.

Epoch AI put six of the most capable large language models through what amounted to a job interview — and none of the candidates got the job. According to the research group's new automation reports, the top models averaged approximately 65% of human-level work quality on real internal tasks.
The research organization's Epoch Automation Reports, released on October 8, 2026, found that no model tested met the quality bar Epoch applies to work completed by its own human employees. The strongest performers landed at roughly two-thirds of the way there. Because the 100% mark is defined by Epoch's own staff and internal tasks, the results gauge a real employer's acceptance bar rather than an abstract test ceiling.
How the Evaluation Worked
Epoch's assessment used open-ended tasks drawn from real work inside the organization, including graphic design and research design. Human graders then scored each output against Epoch's internal standards, with the goal of measuring something closer to actual job performance than a trivia contest. That design choice matters for anyone tracking AI adoption at work: a strong showing on a conventional benchmark does not automatically mean a model's output is something an employer would sign off on.
The scoring scale was anchored to people: a score of 100% represents work of the quality a human employee at Epoch would be expected to produce.
The Scoreboard
Two models shared the top spot, with Claude Fable 5.1 and GPT-6 Astra each averaging approximately 65% across the tasks. At the bottom sat Gemini 3.8 Flash, which scored 42% — leaving a gap of more than 20 percentage points between the leaders and the trailing model.
No model crossed the 100% line. Not one reached parity with the humans whose work set the standard.
The models fared better on structured work, with coding and data analysis standing out as relative strengths. Every model struggled with open-ended tasks that called for judgment or required tailoring work to a particular audience.
The leaderboard has also held steady. As of November 17, 2026, the rankings had not changed since the October release.
Why Epoch Built Another Benchmark
Epoch's automation reports are designed to address shortcomings in conventional performance measurement by reflecting the tasks people actually perform at work. The project sits within Epoch's broader mission of tracking AI progress and capabilities accurately. That framing turns model comparisons into the question employers actually ask: how close is the output to work a hired human would deliver?
What the Findings Mean
Models that look superhuman on structured tests landed well short of human quality once the work required judgment calls. Structured work like coding and data analysis appears to be where current models pull their weight, while tasks that hinge on audience awareness and discretion still seem to need a person in the loop — or at least a person checking the output.
Two rival labs sharing the top score suggests no single developer has pulled decisively ahead on this kind of open-ended work. Meanwhile, the 42% result for Gemini 3.8 Flash illustrates how wide the spread can be among models marketed for serious use.
Between October 8 and November 17, 2026, no model moved up the rankings. The next model release will face a clear and public test: whether it can push past approximately 65% on the kind of work that real employees do every day.
Disclosure: This article is based on reporting by CryptoBriefing, edited by Diego Almada Lopez. Original article: https://cryptobriefing.com/epoch-ai-llms-fail-employee-quality-standard/