Taan launched a music model you can tell what to play, and when

Taan launched on Y Combinator this week with a music model that follows timing instructions in plain English. The product is called Taan v1-preview. The launch page on Y Combinator has 23 upvotes. Taan is in the YC Fall 2026 batch.

Today's music models write catchy tracks. They do not place a cue at a specific second. Ask one to "bring in the violin at the 11th second" and it usually writes a fresh take. The user regenerates until one take fits. That is fine for background music. It is not fine for a video that needs a cue on the frame.

What Taan does

Taan takes a video, an image, a MIDI file, a hummed melody or a script as input. It identifies scene cuts and mood. It composes a score in one pass. The user can direct it in plain English or send no instructions. Taan says every track is original and comes with downloadable stems.

The model is tuned for orchestral and cinematic music today. The company says more genres are on the way. Taan also exposes an API and an MCP server, so a video tool or an agent can call it directly.

The vendor's benchmark

Taan tested four music models on 250 timing prompts. Each prompt asked for a cue at a named second. Some asked for a pause of a named length. Others asked for a mood change at a named moment. Every model was graded with the same detectors and the same tolerance. These are Taan's own numbers, on a test Taan designed.

Taan reports instruction success at 80.1 percent. ElevenLabs reached 28.9 percent on the same set. Google Lyria reached 23.9 percent. Suno reached 22.0 percent.

On timed events landing within ±500ms, Taan reports 73.2 percent. ElevenLabs reported 12.8 percent, Google Lyria 8.6 percent, and Suno 6.8 percent. Taan reports that it left zero events out entirely. The other three models skipped events rather than placing them.

The company has published its full results at taan.ai/research/control-benchmarks. The benchmark was designed, run and graded by Taan. An outside team has not yet reproduced it.

The team

The CEO is Adithya. The company says he comes from eight generations of Indian classical musicians. He has toured as a metal guitarist. He has done music machine learning research since 2023. He has filed two patents on music generation architectures.

The CTO is Devansh. He studied statistics and data science. He did signal processing research at IIT Kanpur. He was an ML engineer at American Express before this company.

What the launch does not say

The launch page does not name a round size, a lead investor, a funding amount, or a customer by name. Taan's own public pricing starts at 15 dollars a month for creators. The next step is a track an outside team can compare, and a benchmark run by a third party.

Source

• Primary source: Y Combinator: Taan.ai launch post, by Adithya.
• Company site: taan.ai.