Sensia Gen 8 · Benchmark report

Signal before model: the Sensia Gen 8 benchmark.

What Gen 8 is, how we tested it on five production analyses, and the full results: selection reach, consistency of findings, resources and limits.

Request the report

What Sensia Gen 8 is The new generation of our qualitative intelligence engine: its five functions, and the role general-purpose language models play in it.
Protocol and portfolios Five production analyses, twenty sections and three portfolios in skincare and pet food, regenerated at six depth configurations.
Full results Selection reach, coverage, main findings, depth of evidence and resources, configuration by configuration.
122 generations25 pages
Limits, stated plainly What the benchmark shows and what it does not, with definitions and sources.

Key results

Four results from the Gen 8 benchmark.

Selection reach

Across four questions, samples of around 1,500 reviews included products accounting for 98.1–99.5% of the relevant review volume. This is not a percentage of insights recovered.
Consistent main findings

The main strategic findings remained consistent across six depth configurations, from a few hundred reviews per section to the writing model's maximum context window.
Depth as a setting

Additional depth expanded the supporting evidence in some analyses. At the maximum window, LLM spend was about thirteen times that of the roughly 1,000-review configuration.
A reasoning configuration

On two sections, a high-reasoning configuration produced comparable main findings, adding nuance in wording at more than thirty times the LLM spend.