Falcon 2.1 key stats
- Latency (measured)
- 117 ms1
- Voice cloning
- Instant cloning, up to 30 seconds of reference audio5
- Released
- October 19, 20266
- Vapi streaming benchmark (50 trials per model) (checked 2026-10-27) Median of 50 sequential Falcon 2.1 WebSocket trials after two discarded warm-ups, arena Clara instant clone, en-US, mono 24 kHz PCM over the global router. Request send after connection open to first audio; includes benchmark-machine network RTT. Zero errors and rate limits.
- help.murf.ai/murf-api-languages (checked 2026-10-27) Count of distinct languages in the Falcon list, excluding regional accent variants. Only English is evaluated by the Index.
- murf.ai/api/docs/text-to-speech-models/falcon-2-1 (checked 2026-10-27) Falcon 2.1 public API rate: $0.01 per 1,000 characters, equivalent to $10 per 1M. Workspace agreements may differ. Instant cloning requires workspace enablement.
- murf.ai/api/docs/text-to-speech/web-sockets (checked 2026-10-27)
- murf.ai/api/docs/voices-styles/instant-voice-cloning (checked 2026-10-27) Requires workspace enablement. Tested with 25-second, 24 kHz mono excerpts of the four licensed source voices.
- murf.ai/api/docs/changelog/2026/10/19 (checked 2026-10-27)
Background
Falcon 2.1 launched on October 19, 2026 as Murf’s streaming speech model. The API accepts the explicit falcon-2.1 identifier over HTTP and WebSocket. Earlier Falcon aliases now route to this generation, so the Index uses the published 2.1 identifier.
Sources: murf.ai
On the Index
The Index generated 80 fixed English clips through the Murf API using four Instant Voice Cloning voices made from licensed source recordings. Every voice reads the same 20 scripts as the other models. Listener votes determine its Humanness score; this scripted evaluation does not measure live turn taking or multilingual performance.
Position in the rankings
Standings as of Oct 27, 2026, 18:46 UTC
Frequently asked questions
- How is Falcon 2.1 tested on the Humanness Index™?
- Listeners hear Falcon 2.1 against another model in a blind head to head round, both voices reading the same customer support prompt from the same cloned source voice, and they pick whichever sounds more human. Its Humanness score derives purely from those votes.
- Which Murf model and voices does the Index test?
- The tested API model is falcon-2.1, using instant clones of Clara, Emma, Godfrey and Nelliot. Gen2 and stock library voices are not part of this result. Murf requires cloning access to be enabled on the workspace.
Find the most human-sounding voice for your agent.
Compare the models in blind tests, read the methodology, or get in touch.
Build a TTS model? Add yours to the Index.
