StepFun’s StepAudio 3 is a family of commercial audio APIs, while Liquid AI’s LFM3-Audio-2B is an open-weight model teams can run themselves. OrcaRouter’s comparison finds the practical choice is less a clean quality contest than a trade-off between hosted features and operational control.
StepFun Watch analysis
What happened
StepFun released five StepAudio 3 models on 15 September: Realtime, ASR, TTS, Gen and Music. OrcaRouter says Realtime supports full-duplex conversation, tool calling and asynchronous tasks. Liquid AI’s LFM3-Audio-2B, published on 27 October, is a 2-billion-parameter checkpoint intended for tasks including recognition, synthesis, translation, audio understanding and spoken tool calls.
The comparison says StepAudio 3 is available as a hosted service, with a limited-time free preview and no announced post-preview price for Realtime. LFM3-Audio-2B provides downloadable weights under the LFM Open License: commercial use is free below $10 million in annual revenue, with paid terms above that threshold, according to OrcaRouter.
Key findings
- Realtime conversation
OrcaRouter says StepAudio 3 Realtime supports interruption and turn-taking; it makes no such duplex claim for LFM3-Audio-2B. - A score with a caveat
StepAudio 3 Realtime scored 98.9% on Artificial Analysis’s Conversational Dynamics board, according to the comparison. That measures conversation dynamics, not a direct head-to-head against Liquid’s model. - Different evidence for different models
OrcaRouter says Liquid’s published figures are vendor-reported and lack a named test set or independent reproduction. - Different operating burdens
StepAudio 3 is a hosted API; LFM3-Audio-2B can be run locally, leaving deployment and maintenance to the operator.
Why it matters
The model that wins a leaderboard may not be the model that fits a product. A team building a voice agent that must handle interruptions may value StepAudio 3’s claimed realtime design. A team needing downloadable weights, local deployment or control over its infrastructure may prefer Liquid’s approach, while taking on the work of running it.
The comparison also flags a practical question for StepFun: the Realtime tier is in a free preview, but its price afterwards is not stated. That makes it a useful moment to test, but a rather awkward time to plan long-term costs.
Our read
This is a comparison of different products, not a verdict that one model is better at everything. OrcaRouter’s account gives buyers useful distinctions, but its published figures should be read in context: some are third-party leaderboard results, others are vendor claims, and the headline scores do not measure the same thing. The sensible next step is to test the model against the actual conversation, languages and deployment constraints you need.
What to watch
- Whether StepFun announces Realtime pricing after the preview.
- Whether Liquid publishes independent evaluations for LFM3-Audio-2B.
- Whether comparable tests emerge for both models on the same audio tasks.
Discussion spark: For a voice product, would you choose a hosted model with realtime conversation features or open weights you can run and control yourself?
Sources and evidence
- LFM3-Audio-2B vs StepAudio 3: The Open 2B Against the Scored Product Line – OrcaRouter (27 October 2026, 19:33 UTC)
not affiliated with or endorsed by StepFun