Decagon's Alto Model Lifts Call Resolution Rates Over Five Points

Decagon says its new model Alto raised resolution rates by more than five percentage points on live customer calls while answering in roughly 700 milliseconds. The company disclosed neither the customers, sample size, baseline rates, nor measurement period behind that figure. The claim is Decagon's own, and the announcement offers no external verification.

Decagon introduced Alto on November 11, 2026, as a Decagon Labs model that controls the conversational layer of its duplex voice agent. Its job is narrow but consequential: deciding when the agent should speak and what it should say. The company said the rest of the agent's logic remained effectively unchanged during the resolution comparison, which at least isolates the variable — on paper.

What problem does Alto actually solve?

A conventional cascade voice pipeline works in sequence: transcribe the caller, reason about a response, then generate speech. Decagon's duplex agent runs two models in parallel instead. A fast conversational model listens and speaks; a more powerful reasoning model handles complex decisions and actions behind the conversation. The agent keeps interacting while the heavy work is underway.

That parallelism forces a judgment call on every turn. Waiting for the reasoning model disrupts the flow; answering too soon risks an unconfirmed statement — telling a caller their reservation has changed before the system has confirmed it. The speed advantage of the smaller model carries a built-in reliability problem that duplex architecture alone does not solve.

Alto addresses this by continuously tracking what the caller has said and what the rest of the agent is doing, grounding responses in a shared call record. It waits for confirmation before announcing an action is complete. During a lookup or tool call, it requests the next piece of information or checks in with the caller rather than falling back on generic filler. If a result arrives while the caller asks something new, Alto can answer the question and deliver the result together.

How was the model trained?

Decagon started with a small open-weight model to keep responses fast, then fine-tuned it on simulated calls informed by production experience. In reinforcement learning, Alto generated several possible responses at each conversation point. The scoring process prioritized whether the model made the correct answer-or-wait choice, then assessed whether its wording followed business policies and reflected confirmed information.

Decagon concentrated training on ambiguous cases after finding that Alto already handled most situations correctly. It also varied policies and tools across simulated businesses, aiming to teach judgment that transfers between situations rather than a memorized script for one company — a reasonable goal, though transfer across businesses is exactly what the undisclosed test conditions make hard to check.

Before any caller heard the model, Decagon reported testing it across more than 90 million simulated calls. That equals roughly one call simulated every second for nearly three years, compressed into a testing pipeline. It then ran Alto in shadow mode alongside live production traffic, where its responses could be evaluated without being spoken to callers.

How does Alto fit Decagon's roadmap?

Alto follows Voice 3 and Chord, announced on October 1, 2026. Voice 3 combined Chord's speech generation with the duplex architecture and support for more than 70 languages. Chord shapes how Decagon's agents sound; Alto shapes how they conduct a conversation. Decagon previously credited Chord with an independent resolution increase of up to seven percentage points — meaning the company has now stacked two separately claimed gains without publishing a combined, audited figure.

Together, the models split voice performance into two training targets: speech naturalness and conversational judgment. Alto's reported result ties the second target to resolved customer calls, but its central operating constraint remains the answer-or-wait decision — using the conversational model's speed without letting a natural-sounding exchange get ahead of confirmed actions. For enterprises weighing deployment, the open question is whether the five-point gain survives conditions Decagon has not described.