# Is Hume Audio Decision Model SOTA?

Pending benchmarks. Announced Nov 2, 2026; no independent results yet.

As of Nov 22, 2026, Hume Audio Decision Model is pending benchmarks on isitsota: announced Nov 2, 2026, with no independent results on the boards we track yet.

Hume AI's audio-native decision model, served through the Hume Audio Decision API. It takes a recording plus questions written in plain language, such as how many speakers there are or whether a caller sounds frustrated, and returns a probability for each answer you define, estimating how human listeners would respond, rather than a transcript or text. API access is by request (waitlist); no weights are mentioned.

Canonical: https://isitsota.com/m/hume-audio-decision-model
Last updated: 2026-11-02
Tracking since: 2026-11-02

## Announced facts

| Fact | Value |
| --- | --- |
| Lab | Hume AI |
| Announced | Nov 2, 2026 |
| API | Waitlist · Hume Audio Decisions API, POST /v0/decisions/file on api.cloud.hume.ai with an X-Hume-Api-Key header; no model parameter, and each response carries the run_id of the model run |
| Open weights | Unknown |
| Type | Audio-native decision model: audio and plain-language questions in, probabilities over the answers you define out; it returns no transcript, reply or explanation |
| Question types | noul (yes/no), choice (pick one of 2 to 20 options), score (2 to 10 ordered levels) and moment (when an event occurs) |
| Answers | Each probability estimates the share of human raters who would give that answer; all questions in a request are answered in one model inference |
| Input | One MP3 or WAV file per request, 5 seconds to 5 minutes and up to 25,000,000 bytes; the model hears audio as one channel at 16 kHz; longer recordings are split into files of 5 minutes or less |
| Request limits | Up to 20 noul, choice and score questions plus up to 5 moment questions per request, optional context up to 1,000 characters, 60 requests per minute per organization |
| Access | By request: Hume's blog says to join the waitlist, and the API returns 403 forbidden for organizations not yet permitted; requests also need a payment method on file |
| Billing | By input tokens (the audio plus the question and context text); answers are not billed. Prices are on the signed-in Hume Platform plan page, not published |
| SDKs | Python hume-audio-decisions and TypeScript @humeai/audio-decisions, both first published Nov 2, 2026 |
| Parameters | Not disclosed |
| Other Hume APIs | Hume is shutting down its EVI speech-to-speech and TTS APIs on Nov 13, 2026, per its voice API changelog |

Tags marked unsure are not stated on the lab's own pages.

## What would let us call it

Verdicts use only independent boards we already track. Two consecutive snapshots are required before any crown.

- [x] Publicly announced with a source
- [ ] Usable by the public (API or weights)
- [ ] Listed on an independent board we track
- [ ] Enough results for a rank in a category
- [ ] Holds across two snapshots (crowns only)

## What the lab claims

Self-reported. Shown for context; not used in any verdict.

- Lab claim: 70.7% agreement with a human panel's average judgment in an initial evaluation, against 56.5% for an individual human rater.
- Lab claim: an average of 0.82 across 10 audio benchmarks at 0.23 seconds median latency per screen, against 78% for Gemini 3.8 Flash, 76% for Qwen3.8-Omni-Flash and 67% for Inkling-Small, with gpt-audio-1.5, Gemini 3.5 Flash Lite and Nemotron 3 Nano Omni also compared.
- Lab claim: trained on more than 10 million human judgments of how people perceive speech and sound.
- Lab claim: replaces what used to take 10+ models with one call, and is fast enough to ask dozens of questions every few seconds on a live call or make an archive of 40,000 recordings searchable in minutes.
- Lab claim: the first audio-native model that answers questions about speech and sound.

## Sources

- LAB: [Introducing Hume's Audio Decision API (Hume blog)](https://www.hume.ai/blog/introducing-hume-audio-decision-api) (2026-11-02)
- LAB: [Hume AI announcement on X](https://x.com/hume_ai/status/2117285608745820372) (2026-11-02)
- LAB: [Andrew Ettinger (Hume AI CEO) post on X](https://x.com/ahe23/status/2117285610071134317) (2026-11-02)
- LAB: [Audio Decision API product page](https://www.hume.ai/products/decision-model-api)
- LAB: [Audio Decisions API overview (Hume docs)](https://dev.hume.ai/audio-decisions/docs/overview)
- LAB: [Audio Decisions API limits (Hume docs)](https://dev.hume.ai/audio-decisions/docs/limits)
- LAB: [Audio Decisions API billing (Hume docs)](https://dev.hume.ai/audio-decisions/docs/billing)
- LAB: [Audio Decisions API changelog](https://dev.hume.ai/audio-decisions/changelog) (2026-11-02)
- LAB: [Hume voice API changelog (EVI and TTS sunset)](https://dev.hume.ai/voice/changelog)

Tracking ends Jan 31, 2027 unless the API or weights become available.

## Related

- Lab: [Hume AI](https://isitsota.com/labs/hume-ai.md)
- All pending releases: https://isitsota.com/releases.md
- JSON: https://api.isitsota.com/v1/models/hume-audio-decision-model
