Latency pushed toward the sub second range

Watch a match and the neighbour cheers a few seconds before your phone catches up. That gap is not a bug in your connection, it is the price of the delivery model that most live video grew up on: segment the stream into chunks, ship the chunks, let the player buffer enough to survive jitter. It is robust, it scales to millions, and it makes genuine interaction impossible. Ant Media Server attacks exactly that gap. Playback over WebRTC lands around half a second, which is the range where a conversation feels like a conversation rather than a radio call.

The scenarios that need it are specific. Live auctions where the bid must land in the same second for everyone. Online classes where a student raises a hand and expects an answer, not a delayed reaction. Any format where the audience is supposed to respond to what they see. For those, a self hosted option that reaches sub second delivery has been the missing piece.

Under the surface it is a Java streaming engine, and the community edition is open source under the Apache licence. The chain is complete rather than partial: ingest, transcoding, and distribution across several protocols come in the same product, so there is no need to stitch three projects together with a message queue and hope the clocks agree. Docker deployment means a small team can stand the whole thing up in an afternoon instead of a quarter.

3.1 reaches into professional production

The 3.1 release, shipped in September 2026 under the tag 2026-09, points its attention at production environments. NDI sources can now be ingested, so professional devices in a studio, a classroom or an enterprise network are discovered automatically and flow straight into the live pipeline. SRT ingest supports multiple audio tracks with language tags, which settles the distribution question for simultaneous interpretation and alternate commentary. The admin console has been rebuilt into a more modern interface, currently in beta, the runtime has moved to Java 21, and GPU encoding gained better load balancing across several cards.

The two protocols are not in competition here, they divide the work. HLS handles playback for anyone watching in the traditional way, with the buffering and the compatibility that implies. WebRTC handles the interactive audience that cannot tolerate the delay. Choosing per session rather than per product is what a real deployment looks like.

The economic argument is worth spelling out, because self hosting is often dismissed as a hobby. A managed live platform charges by minutes delivered plus a per viewer fee, and the meter runs whether the room is full or empty. A server you own charges by bandwidth and by your own time. For an organisation that already runs infrastructure, the second bill is familiar and predictable; for one that has never operated anything, the first is cheaper in every sense. The honest test is whether you can answer a page at two in the morning when a stream fails.

What makes the low latency claim credible rather than aspirational is the shape of the delivery. Interactive sessions do not need the segments and buffers that make HLS reliable for a mass audience, they need a path short enough that a hand raised on camera is answered while the hand is still up. Keeping both paths in the same product, and choosing per session, is the design decision that makes the numbers real in practice.

This is the solid route for open infrastructure: take the hard problem of low latency live video and work through each link, ingest, transcoding, distribution, playback, until every one of them holds up. Professional capability in open code, and small teams are the biggest winners.

A word on selection. The community edition is functionally complete, while clustering and some advanced capabilities belong to the enterprise edition. Start with a single node on the community build and talk about horizontal scaling once the volume arrives, because the architecture you need at ten viewers is not the one you need at ten thousand. Bandwidth is the real cost centre in live video, so measure it before promising a price to anyone.

For teams that need to own their delivery chain, online education platforms, interactive events and camera aggregation among them, it is the self hosted option with the clearest value. You trade a monthly bill for an operational responsibility, and for the right team that is a good trade.