Quick summary: The X2 announcement replaced the experimental X1 architecture and was presented by Secrets AI as a major intelligence improvement. Those statements are developer claims from the release note, not independent performance findings.
The release replaced X1 rather than adding a permanent menu item
v5.4.0 introduced X2 and marked X1 as deprecated. The changelog said X2 used a different architecture and should compare with or exceed S3.5 performance. It also paired the release with memory and long-term recall improvements plus a caching fix.
A historical model name should not be assumed to remain selectable. Secrets AI has repeatedly consolidated, removed, or migrated chat modes. The current account interface and Chat Features documentation are more reliable than an old release name for deciding what can be selected today.
AI Video Gallery
A preview of selected portrait videos and AI character images.
Performance language needs careful attribution
Phrases such as “exceeds” or “more intelligent” came from the product release. SecretOpedia has not independently benchmarked the model across a controlled test set, memory depth, safety behavior, response time, and long conversations. A responsible analysis therefore records the claim without turning it into an editorial conclusion.
A useful comparison changes one variable at a time
- Use the same companion and a new or duplicated scenario.
- Keep reply format and Spice Level unchanged.
- Test factual instruction following, role consistency, correction handling, and repetition separately.
- Run enough turns to observe context behavior rather than judging one reply.
- Record which mode was actually shown in the account at the time of testing.
The control relationships are explained in the current workflow for chat mode, Spice Level, and reply formatting.
Memory improvements should not be credited only to the model
The same release included memory architecture and caching work. Better continuity after February 18 could therefore come from model behavior, memory retrieval, or both. the consolidated analysis of the 2026 memory changes separates those layers.
How to read the update now
Treat X2 as a documented stage in the product’s chat architecture. Do not tell users to select X2 unless it appears in their account, and do not assume a later mode is merely a renamed X2. the Secrets AI review’s current conversation-quality assessment should use present controls, not historical release marketing. SecretOpedia’s independent Secrets AI model-update archive preserves the timeline.
A fair user test should keep the companion, chat history, reply format, and Spice Level constant while changing only the available model choice. Comparing unrelated chats can make a model look better or worse because memory scope, accumulated context, and roleplay direction differ. This is an editorial comparison method, not an official benchmark.
The release also arrived alongside major memory work, so improvements observed at the time may not belong to the model alone. The separate memory analysis should be used to distinguish system-level recall changes from the developer’s X2 claims.
The article is marked partially superseded because current public guidance focuses on the controls available in the live chat interface rather than presenting X2 as a permanent label. Users should not assume an older model name will remain selectable after later architecture changes.
Official sources checked July 12, 2026: Changelog, Chat Features, Tips & Tricks, Memory System.
