PRME 1.3 includes an updated AI engine to support more involved technical work. The update focuses on coding, following complex instructions, and handling tasks that require multiple steps.
The engine update was deployed in August 2026. Since 3 September, the model has used the name PRME 1.3 under IsonAI’s new model naming scheme. The naming change and the capability update therefore took place at different times.
Four areas of improvement
Published evaluations of the underlying AI engine report gains over its previous generation in four areas. These inform our description of the PRME update; results in IsonAI still depend on the task and available features.
- Coding: software fixes and work across larger code projects.
- Instruction following: adherence to directions, including tasks with several requirements.
- Multi-step work: planning and adapting actions to results or feedback during a task.
- Visual understanding: reasoning about images, reading scientific charts, and understanding documents.
These areas are relevant to reviewing code, producing a specified format, organizing technical work, or discussing an attached chart. Feature availability follows the IsonAI service.
What have we checked?
Before deployment, the update passed internal checks covering code generation, Indonesian writing, image understanding, and tool use. These checks established that the tested functions worked after the engine change.
These were initial functional checks, rather than a comprehensive measurement of quality gains across all tasks. We therefore do not attach a percentage improvement or promise that every response will be faster or always correct. Results still need to be checked against the work at hand.
Tasks to try
To assess PRME’s usefulness for your own work, provide context and clear acceptance criteria. For example:
Review the following function. Find the cause of its failure on empty input, propose a fix without changing the function name or output format, and write tests for empty input, normal input, and duplicate data. Explain your assumptions.
To explore a technical problem before writing code:
Help me plan a search feature for an order list. Separate user needs, search rules, the no-results state, and acceptance criteria. Flag information I still need to decide.
These are suggested uses, not benchmark results or before-and-after comparisons. For code, run tests in your development environment before applying changes.
Choose PRME 1.3 in IsonAI and try a task you know well. Feedback about missed instructions or incorrect results helps us identify what to improve next.