AI Framework Picks Perovskite Additives Faster
LEAP prioritizes solar-cell additives with a language model and active learning, a research-stage method tested on only a few candidates so far.
Updated on Nov. 4, 2026 in Chemistry

Researchers introduced LEAP, a closed-loop framework that pairs a domain-specialized language model with active learning to find perovskite precursor additives. The work is research-stage, published early as an accepted peer-reviewed article, and only a limited number of candidates were tested prospectively.
Why it matters
Conventional trial-and-error screening struggles because the chemical space of possible additives is very large, so a method that prioritizes candidates under low-data conditions targets a real bottleneck in perovskite solar cell development.
The study measures average power conversion efficiency, the share of sunlight converted to electricity, for two additives against a control device at 19.25%, and reports a best device at 21.32%. The evidence comes from a small set of prospectively tested candidates, and the authors' generalization beyond those cases is not yet demonstrated.
The details
LEAP uses a language model specialized for this chemistry domain to read the existing literature and extract the mechanisms that matter for perovskite precursor additives. That extracted knowledge is then converted into interpretable molecular descriptors, numerical summaries of a molecule's properties. An uncertainty-aware Bayesian optimization step, a method that picks the next experiment by weighing expected gain against model uncertainty, chooses which candidates to test, and the loop repeats as new results arrive.
Timeline
Article published early as accepted peer-reviewed research on 4 November 2026.
The Tech Race
LEAP sits within the broader effort to apply machine learning to materials discovery, with its contribution being a domain-specialized language model paired with active learning rather than general-purpose models. The reported comparison is against tested general-purpose models on an evaluation drawn from unseen literature, which places the framework relative to those alternatives.
LEAP is a research-stage framework with no announced product, pricing, or availability timeline, so it is not yet usable outside the laboratory. Any benefit to solar cell manufacturers would depend on whether the approach generalizes beyond the limited set of candidates tested so far.
The takeaway
The trajectory to watch is whether LEAP's prioritization holds up when tested on a much larger set of prospective candidates. A follow-up study reporting prospective results across more additives would show whether the reported efficiency gains generalize.
Further reading
Explore more on the Chemistry research and developments shaping materials science today.
Source note: This article includes information reported by Nature.


