After the release of Aia - Ai Assistant 2.0 last week, which added support for Anthropic Claude and Google Gemini platforms, it became clear after testing and from user feedback that further prompt and configuration optimisations were necessary—particularly with Google Gemini, which can consume far more response tokens than expected. Since the Spellcheck feature doesn’t require creativity, but rather accuracy and speed, we’ve refined the way requests are structured to keep responses lean and consistent.
Key Improvements
Smarter Spellcheck Configurations
Spellchecking now uses dedicated configurations per platform (OpenAI, Claude, Gemini). This ensures each provider’s quirks are accounted for, while still returning clean, structured results.
Model Choice for Spellcheck
You can now choose a separate model just for spellchecking, independent of your main model. By default, Ai - Ai Assistant will select a faster, more cost-effective model optimised for this task, which means spellchecking is not only quicker but also more accurate.
Extended Model Support
For those who preferred them, we’ve added back the legacy gpt-4o and gpt-4o-mini models from OpenAI. These strike a great balance of speed, cost, and reliability.
Higher Token Limits
The default max_tokens value has been increased to 2048 across the board, with Gemini bumped to 4096 to better handle its tendency toward longer outputs.
Prompt Optimisation
We’ve restructured the general prompt format to reduce unnecessary token usage and to ensure models respond more efficiently. This means less bloat in the background, and faster, cheaper results for you.
A full changelog is available to view here
Thank you to everyone who submitted bug reports and tested development versions. If you find any more issues please submit them on the forum or on github.
Download and Installation
Aia is available as a seperate download and requires a valid JCE Pro Subscription.
Instructions for installing and configuring Aia - Ai Assistant are available here