Sparrow-2
Real-time AI model for natural conversational turn-taking.
At a glance
Price not confirmed · Source review 2026-11-02- Price & billing
Official demo page and Tavus blog reviewed 2026-11-02: the demo invites visitors to talk to a PAL and says “Ready to build with it?”; no standalone plan or amount was verified on the pages reviewed. Recheck official pricing before purchase.
- Free allowance
Not separately confirmed. Check the current plan before relying on a free allowance.
- Best suited to
Voice-agent builders who need natural turn-taking, interruption and backchannel handling in real rooms
Fit is editorial guidance, not a measured result.- Platforms
Access: not separately verified. Operating systems not separately verified.
Published-source facts and editorial guidance. No product test is implied. Confirm current conditions before purchase.
No product screenshot is claimed for this listing. Use the official website ↗ to inspect the current interface.
What to know before choosing
Good fit
Voice-agent builders who need natural turn-taking, interruption and backchannel handling in real rooms
Limits and trade-offs
TEXT-VERIFIED: verified by official-page text fetch on 2026-11-02; not a live-browser render, and not labelled as one. A conversational model with a live demo, not a standalone consumer app; no standalone plan or amount was verified on the pages reviewed. TurnBench figures are vendor-reported preliminary dev-split results. No authenticated product test was run.
Plans and free allowances
Official demo page and Tavus blog reviewed 2026-11-02: the demo invites visitors to talk to a PAL and says “Ready to build with it?”; no standalone plan or amount was verified on the pages reviewed. Recheck official pricing before purchase.
Evidence
Official live demo ↗ (reviewed 2026-11-02), Official blog ↗ (reviewed 2026-11-02)
TEXT-VERIFIED: verified by official-page text fetch on 2026-11-02; not a live-browser render, and not labelled as one.
Dates: Last reviewed 2026-11-02. Last reviewed is the latest source or editorial review date; it is not a content-change date.
A closer look at Sparrow-2
Sparrow-2 is Tavus's audio-native, streaming-first conversational understanding model that jointly models turn-taking, interruptions, backchannels and the full acoustic scene at a native 10 ms frame rate. Its live demo lets you talk to a PAL and watch it track turn-taking, room noise and interruptions as the conversation happens.
Editorial guidance based on published documentation. The workflow below is a suggested starting point, not a measured product result.
When to choose Sparrow-2
- Voice-agent builders who need natural turn-taking in noisy or shared rooms, not just quiet one-speaker calls
- Teams building Tavus PALs and face-to-face conversational agents
- Builders whose systems today rely on silence timers and misfire on mid-thought pauses
- Developers who need backchannels and genuine interruptions told apart instead of treated as the same event
When to choose something else
- Users looking for a ready-made consumer app rather than a conversational model with a live demo
- Users who need a text-only chatbot rather than an audio-native, streaming-first conversational model tracking turn-taking, interruptions, backchannels and the acoustic scene at a 10 ms frame rate.
A practical starting workflow
- Open the Sparrow-2 live demo and talk to a PAL; your microphone is what matters, and the camera can stay off.
- Speak naturally, including mid-thought pauses, and watch the demo track turn-taking states such as still speaking, acknowledging and cutting in.
- Let the model sort the room: it distinguishes speech directed at the PAL from other voices and room noise, and counts backchannels, noises and interruptions.
- Test the hard cases the model is built for: a backchannel such as “mhm” while the PAL speaks, a genuine interruption, and unclear audio it should not answer confidently.
- When you are ready to build with it, follow the demo's build path into the Tavus platform rather than treating the demo page itself as the product.
Check before committing
- TEXT-VERIFIED: verified by official-page text fetch on 2026-11-02; not a live-browser render, and not labelled as one.
- Sparrow-2 is a model that powers Tavus PAL conversations, not a standalone consumer app; no standalone plan or amount was verified on the pages reviewed.
- The TurnBench figures quoted here are vendor-reported and preliminary: the vendor's own overview says they are preliminary results on TurnBench's public dev split, with full test-set results still to be published.
- Vendor claims on the official pages are not guarantees; actual timing, accuracy and availability must be tested in the product.
- Prices and features may change after this review; recheck the official site before purchase or signup.
Common questions
What is Sparrow-2?
Sparrow-2 is an audio-native, streaming-first, multilingual conversational understanding model from Tavus that preserves the full audio scene and lets conversational decisions, such as when to listen, wait, speak or continue speaking, emerge from that understanding.
How is Sparrow-2 different from silence-timer turn detection?
Instead of waiting on a fixed silence threshold, it models semantics, lexical structure, prosody, timing and speaker identity together, and it holds through genuine mid-thought pauses of up to 6 to 8 seconds, where prior systems were typically limited to 1 or 2.
How does Sparrow-2 handle interruptions and backchannels?
An “mhm” while the PAL is speaking can be understood as a backchannel telling it to continue, while a genuine interruption can tell it to stop; the demo separately tracks backchannels, noises and interruptions.
What is Sparrow-2's TurnBench standing?
On TurnBench, the independent benchmark of real human conversation, it leads all fourteen published systems in both end-of-turn detection (92.4% recall) and interruption detection (97.4% recall); the vendor notes these are preliminary results on the public dev split, with full test-set results still to be published.
How can I try Sparrow-2?
Talk to a PAL on the live demo page, where your microphone is what matters; the demo shows turn-taking, room noise and interruption states live, and ends with an invitation to build with it.
Compare approaches for this task
Sources and checks
Last source review is not a date of product testing. No rating or test result is implied by inclusion. Check the current plan and terms before buying.