Version 2.22.0
GPT-5.5, custom skills for assistants, Slides on by default
What's new
- GPT-5.5 available, GPT-5.4 in EU hosting — The new top model GPT-5.5 is available in the model picker. GPT-5.4 is now also served from European data centres.
- GPT-Image-2 for image generation — The new GPT-Image-2 image model is available in the image generator with improved detail.
- Custom skills for assistants — You can now equip assistants with your own reusable skills, encapsulating common workflows as building blocks.
- Slides on by default — Slides is now enabled by default for activated workspaces; a first-run explainer modal walks you through the main features.
- Clearer tool display in chat — Tool calls and their results are presented more clearly — what's running, what came back, and what happens next.
- Tidier settings tables — Settings tables were modernised with more consistent sorting, filters, and faster rendering.
For administrators
- Image-token usage tracking — Token usage from image generation is now captured fully and shows up in cost reports.
- Optimised LLM caching — Repeated requests use less context and cost less; especially noticeable in long conversations with shared context.
- No more 10-invite limit on pilot accounts — The 10-invitation cap was removed; pilot customers can invite larger groups in a single step.
Fixes
- Advanced web-search settings save again — Settings for the advanced web-search tool persist reliably again.
- Confluence connector in assistants — Active Confluence connections are fully usable from assistants again.
- Disabled models blocked via API — Models disabled in org configuration can no longer be used through the API.
- More robust PDF processing — Broken PDFs block extraction less often; an additional fallback handles malformed PDF structures.