Meet Cara-4.1 Express, our most expressive realtime avatar model
Today we’re releasing Cara-4.1 Express, our latest real-time avatar model. It brings more expressive animation, better lip sync and closer control over how an avatar performs. Alongside it, we’re reducing the time it takes avatars to connect and reply, and adding support for much wider framing of custom avatars. In blind human evaluations, 85% of raters scored Anam higher than HeyGen, 88% higher than Tavus and 92% higher than LemonSlice.
Human evaluations
In a blind study, sixty people rated Anam, Tavus, HeyGen and LemonSlice for visual quality, naturalness, lip sync, interruptibility, response time and overall experience.
Average human rating across six questions, on a −2 to +2 scale. Error bars show 95% confidence intervals.
Based on each person’s average scores across four conversations with each provider, 51 of 60 raters scored Anam higher than HeyGen, 53 higher than Tavus and 55 higher than LemonSlice. Anam also had the highest average score on all six questions, with the clearest advantages in lip sync, naturalness and overall experience.
Share of ratings above neutral. Models: Anam Cara-4.1 Express, Tavus Phoenix 4.5, HeyGen and LemonSlice (default model). Blind study: 60 raters, 960 conversations, 16 October 2026.
Synthesia’s real-time avatars are available only via LiveKit, so we compared against Synthesia and other providers in a separate LiveKit study, using the same speech recognition, language model and voice. Based on each person’s average scores, 42 of 60 raters scored Anam higher than Synthesia, with three tied.
Share of ratings above neutral, across 1,200 conversations.
We’ve published the code so you can reproduce these evaluations.
Director Notes and custom styles
My favourite change in this release is Director Notes coming out of beta, with better prompt adherence and support for custom style prompts for emotion, expression, gaze and head movement. Try “lively, with energetic head nodding and smiling,” or “distracted and bored, with eyes drifting off camera” then shift the mood mid-session with cues like [curious] or [angry] as the avatar speaks. I’m excited to see what people build with this!
Wider avatar framing
Wider framing, one of our long-requested features, is now available for custom avatars in Anam Lab and the API. In Lab, the framing control lets you show more of the character and the room or setting around them.
Zooming out by 40% to fit a 9:16 mobile screen.
Faster connections and faster replies
Quick connections let people get straight into a conversation, while fast replies keep it flowing naturally.
In client-side tests across multiple regions, 95% of Anam sessions connected within 2.97 seconds. That’s faster than the median connection time of the next-fastest provider in the study. Anam also had the fastest median response time at 1.24 seconds, measured from the end of the user’s speech to the start of the avatar’s reply.
926 sessions and 4,703 turns. Further left is faster. The response-time bump near zero is an artefact of how interruption latencies are measured.
Try Cara-4.1 Express
Cara-4.1 Express is available now in Anam Lab and through the API. Set avatarModel: "cara-4.1-express" in your persona config.
We’re already working on Cara 5. More on that soon. 👋
Explore more articles
© 2026 Anam Labs
SOC 2 Type II report · HIPAA-aligned controls.