Opus is the universal-default audio codec for 2026 — WebRTC, Discord, podcasting, streaming. The defaults are good; the knobs you actually turn are bitrate, complexity, and FEC. Each has a specific use case where it earns its operational cost.
Bitrate by use case
Voice (mono): 24-32 kbps is transparent for most listeners. Music (stereo): 96 kbps is the perceptual threshold. Premium podcast: 128 kbps stereo. Below 16 kbps voice becomes warbly. There's no reason to go above 192 kbps for any human listener.
Complexity dial
0-10 scale. 5 is sane default. 10 = best quality, ~3x CPU. 0 = fastest, audible quality drop. Real-time encoders (WebRTC) often use 5; on-demand encoders use 10. Decoding is fixed-cost regardless.
Forward Error Correction (FEC)
Embeds redundant data in the next packet so a single packet loss doesn't cause a gap. Costs ~5-15% bitrate. Worth it for VoIP over unreliable networks; redundant for HLS where retransmission exists.