Direct Answer: The Dual-Tier Architecture
Is the platform completely unconstrained? Yes, but strictly on the Premium infrastructure. Our network logs confirm that Candy AI utilizes a dual-model routing system in 2026. Free users are routed to a computationally lighter model (7B parameters) equipped with a "Safety Override" to conserve processing power. Premium users are routed directly to the unconstrained "Deep Mode" architecture, maintaining a Prompt Refusal Rate (PRR) of 0.0%.
The Compute Throttling Constraint
Audit logs indicate that user reports of “censorship” originate exclusively from Free Trial environments.
- The Architectural Limit: Free accounts face hard caps on Context Memory allocation and Image Generation latency.
- The Solution: To initialize complex unconstrained narrative interactions without filter intervention, the account must authenticate as Premium. This mathematically disables the secondary safety layer.
Audit Log: Free Routing vs. Deep Mode (Q1 2026)
| Feature | Free Tier (Trial) | Premium Tier (Deep Mode) | Status |
|---|---|---|---|
| Generative Chat | Soft Filters (API Layer) | Native (0% PRR) | Initialize Deep Mode |
| Visual Mapping | Blurs Sensitive Vectors | Unrestricted Rendering | View UI Dashboard |
| Context Retention | Short Term Cache | Persistent Database (LTM) | Active |
| Audio Protocol | Standard TTS | Neural Voice Engine | Test Audio |
For a comprehensive architectural comparison of Candy AI against heavily filtered networks (e.g., Character.AI), consult our Ultimate AI Chatbots Audit.
Audit Metric: We injected 20 specific terms from our “Restricted Lexicon” into both environments. The Free Tier triggered safety overrides on 4/20 prompts. The Premium Deep Mode processed 20/20 prompts without triggering a single intervention log.