The New Benchmark: CLARISSA + Opus 5.5
CLARISSA · · Opinion
AI reservoir simulation has changed quickly.
A powerful frontier model such as Claude Opus 5.5 can increasingly read technical material, interact with software, create simulation inputs, and run tools such as OPM Flow.
So simply saying “our AI can run a reservoir simulator” is no longer enough.
The new question is:
Can the AI be trusted?
That is where CLARISSA is designed to be different.
A simulation can run successfully and still contain poor engineering assumptions. Relative permeability, saturation endpoints, PVT, well behavior, injectivity, and numerous other decisions can materially affect the answer.
CLARISSA is being built with reservoir-specific verification around the underlying AI: physics checks, engineering calculations, simulator integration, and a framework for evaluating whether results are reasonable and reproducible.
It is also model-independent.
When a stronger model appears, CLARISSA does not have to compete against it. That model can become part of CLARISSA. The current architecture is designed to support different models, including frontier systems such as Opus.
That gives us a straightforward test:
Give raw Opus + OPM Flow and Opus inside CLARISSA the same reservoir problem.
Then compare:
- engineering errors,
- simulator failures,
- error detection,
- recovery,
- reproducibility,
- runtime,
- and required human intervention.
The CLARISSA team has identified this same baseline internally: as raw Claude plus OPM Flow becomes increasingly capable, CLARISSA must prove additional value through better capability or better usability.
That is the benchmark we should welcome.
CLARISSA does not need to be smarter than Opus 5.5.
It needs to make Opus 5.5 better at reservoir engineering.