Pika launches a full-song music model for its API

Pika has launched an audio-model family covering music, sound effects, speech and soundtrack generation. Pika Music is the full-song member: the company says it can produce a complete track of up to six minutes from a text prompt and lyrics, with optional musical and voice references.

The API exposes text-to-music, voice-conditioned text-to-music and music-cover modes. A voice reference requires the user to confirm consent to use the sample. That is a necessary product control, although a consent checkbox is not independent verification of identity or authority.

The pricing changes the experimentation loop

The live price is USD 0.015 per generated minute, charged for successful generations. At that rate, a four-minute output costs USD 0.06 and a six-minute output USD 0.09, before any separate API Club membership cost. Pika reports that a 90-second song takes 6.21 seconds on average in its own testing, or about 14.5 times faster than real time. Both speed and comparative-cost claims are vendor benchmarks until reproduced under matched conditions.

The current output specification lists MP3. Access is through the Pika API Club and its web playground rather than Pika's main consumer video editor.

The strategic signal is clear even before a broad listening test: fast, inexpensive full-song generation is becoming an API primitive. That makes iteration cheap enough for applications to generate several candidates, rank them and regenerate around a reference instead of treating each song as a costly one-shot request.