Salesforce AI Research published details of Launa-Omni, a voice model for real-time, two-way conversations, on October 19, 2026. According to the Salesforce blog post, the model "was developed and trained for experiences in Salesforce and Slack." Salesforce has not announced availability, an API or pricing for Launa-Omni.

#What is Launa-Omni?

Launa-Omni is a voice model that pairs an audio-adapted text model with a separate speech generator. "Its core is a text model retrofitted for audio, so speech and language share a latent space," Salesforce wrote. "Outside that core model, an in-house-trained text-to-speech (TTS) model produces the spoken reply."

Salesforce describes three design goals:

  • Text first. "Launa-Omni produces a text response before anything is spoken, so the reply is visible at the moment it is formed and teams can inspect and correct it."
  • Customization. "Launa-Omni can be customized and fine-tuned to better handle named entities, dates, alphanumeric IDs, and industry jargon."
  • Turn-taking. Salesforce says the model's real-time harness "detects speech and determines when a person’s turn is complete."

Salesforce also says "The core audio adopted text model is significantly smaller than other leading voice models," but gives no size figure. It says Launa-Omni "can be paired with a frontier model behind the experience" for requests that need longer reasoning or search.

#How does it differ from other voice systems?

Salesforce positions Launa-Omni between two common designs. In its blog, it says cascaded voice systems, which chain separate speech recognition, language and speech generation models, "can lose details from the original voice and add pauses." It says speech-to-speech models can bring "higher serving cost and less visibility into how a reply was formed."

Launa-Omni's answer, per Salesforce, is to keep the text response visible while handling speech inside one audio-adapted model. That is Salesforce's framing of the tradeoffs, not an independent comparison.

#What do Salesforce's benchmarks show?

All of these scores are Salesforce's own. Where the blog and the Launa model card differ, this article uses the blog's figures.

Benchmark

Launa-Omni score (Salesforce's claim)

Note from Salesforce

VoiceBench

90.60%

Same checkpoint as Audio MultiChallenge

Audio MultiChallenge

59.69%

Average across three runs; top public score shown is 60.40%

τ³-Voice

72.9%

Macro score with a custom harness, on customer-service tasks in retail, airline and telecom

The third-party VoiceBench leaderboard lists "Launa V1 (90.6)" at number one overall. The leaderboard notes that "Launa V1 scores are from the authors' updated submission," so the 90.6 figure was submitted by Salesforce rather than measured independently. VoiceBench files the model under its "Audio-LLM" category.

The model card, titled "Launa: Sovereign AI Voice Models from Salesforce," introduces "Launa Omni v1 and Launa Transcribe v2" and says "Launa was trained in partnership with Arcanum." Its comparisons against other companies' voice models are marked "Scores reported by the Launa team."

#Who can use it and when?

Salesforce has not announced availability, an API or pricing for Launa-Omni. Neither the blog nor the model card gives a release date.

The only version in use today, according to Salesforce, is a simpler one: "Slack users can already experience a simplified version of this model, called Steno, in its speech-to-text dictation feature, Slackbot Voice Dictation."

Item

Status, per Salesforce

Launa-Omni availability

Not announced

Launa-Omni API

Not announced

Launa-Omni pricing

Not announced

Steno in Slackbot Voice Dictation

Live, a simplified version of the model

#How does it compare?

Salesforce says Launa-Omni targets experiences in Salesforce and Slack. For how other AI agents work inside Slack, see our guide to OpenAI Dots in Slack and Microsoft Teams. For another enterprise agent platform move, see our coverage of specialist OpenAI Dots for Microsoft Agent 365. More voice coverage is in the Voice Agents hub.

#What is still unclear?

  • Availability and pricing. Salesforce has not said when or how customers can use Launa-Omni.
  • Model size. Salesforce calls the core model significantly smaller than other leading voice models but gives no number.
  • Model card terms. The model card does not explain "Sovereign" or Arcanum's role.
  • Independent results. No independent evaluation is in the sources; the VoiceBench listing reflects Salesforce's submission.