Flash is now the default model

Our fastest model now powers every avatar on every plan, with no code changes needed.

Flash, our fastest model, is now the default for every avatar on every plan. Every session gets Flash's lower latency — no configuration change required.

What changes

Flash responds about 200–300ms faster than our previous default, LemonSlice 2.1. In our April 2026 benchmarks, it reached a 471ms time-to-first-byte and ~2s end-to-end response times when paired with third-party speech-to-text, LLM and text-to-speech. Faster LLM and speech models available today reduce perceived latency even further.

Flash used to be limited to Ultra and Enterprise plans and to 2:3 avatars. It now runs on every plan, in every aspect ratio, at 24 fps.

Do I need to change anything?

No. Sessions that don't set model now run Flash. If your code passes model: "flash", it keeps working and gets the same default model, so you can remove it whenever it's convenient.

Ready to build with real-time interactive avatars?