What we checked
Public demo access and published sample playback were checked in Chrome. No live microphone session or controlled cross-provider latency experiment was completed.
Checked October 13, 2026. No credits or subscriptions were purchased for this walkthrough. These are access observations; measured latency, delivered FPS, and cross-provider quality results remain unfilled.
Anam demo
Playground controls inspected. Cara-4 label on the playground.
The ready playground exposed avatar, voice, prompt, LLM, and tool tabs. Opening the transcript exposed a message field that stayed disabled until Start Call. The page's SDK example enables microphone input by default.
No live call started. Ready and avatar-choice controls establish interface access, not returned video quality or response latency.
LemonSlice demo
Live choices inspected; recorded sample played. CWM-1 label on the research page.
After choosing necessary cookies, the page exposed seven stock-character Start call controls. We played the vendor's Conversation with a CWM-1 avatar recording and observed the playback timer advancing with video ready.
The recording is a vendor-provided sample. Its playback does not establish the latency, frame rate, or reliability of a newly generated API session.
Simli demo
Character picker inspected; setup did not advance. Not exposed in the selected demo setup.
We chose Chad from the character picker and entered the test name Demo. Selecting Next left the interface on the same name step in this check. No avatar session or model-specific output was verified.
This single setup observation does not establish a provider reliability rate or the performance of its API.
Tavus demo
Homepage demo entry inspected. Charlie's exact deployed model not exposed.
The homepage exposed Speak with Charlie and a separate Griffin research announcement. We did not start the conversation.
The demo's exact model is unknown. Griffin is a restricted preview, separately from the publicly available Phoenix, Raven, and Sparrow stack.
Make your own demo check
- Select a disclosed stock avatar and note the displayed model and date.
- Keep text, prerecorded audio, and microphone conversations in separate cohorts.
- Record signup requirements, input controls, avatar preparation, session readiness, and any blocked step.
- Use the same permitted inputs and instrument the same playback events before publishing latency comparisons.
- Keep original observations separate from vendor recordings and reported timing badges.
- Record charges cumulatively and stop before the total approved budget is exceeded.
Official sources
Continue your evaluation
Read benchmark clocks and existing evidence · Read the methodology · Create a shortlist