Hume AI is live · free plan, paid from $3/mo · last verified 5 Nov 2026 · viability healthy
Hume AI
Hume AI measures 600+ dimensions of vocal and facial expression to benchmark and improve voice AI agents.
Hume is the vendor to call when your agent passes the accuracy tests but callers still say it feels off, and the free VoiceEQ leaderboards are a sensible first stop before you commit budget. The human-rated scoring layer is what separates it from LLM-judge eval stacks. Skip it if you want voice generation or a turnkey agent platform — TTS and EVI are gone as of November 13, 2026.
Verdict generatedVerified 8d ago · liveness 88/100 · cite: rightaichoice.com/tools/hume-ai
- Voice AI teams whose agents pass accuracy tests but feel wrong to real callers
- Model developers comparing checkpoints and diagnosing weaknesses before release
- Teams that want human-rated voice quality scores alongside automated LLM-judge results
- Companies running private evaluations of TTS, speech-to-speech, or ASR systems
- Teams shopping for voice generation — TTS and EVI access ends November 13, 2026
- Buyers wanting a turnkey voice agent platform with telephony and call handling
- Developers who need a fully open-source voice stack they can self-host and fork
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip Hume if you need a voice model to generate with rather than a way to measure one — the TTS and EVI APIs are sunsetting November 13, 2026, and the value here is evaluation, not synthesis.
Additional TTS characters are billed past your included allowance at $0.15, $0.12, $0.10, or $0.05 per 1,000 characters depending on tier, so a long-running narration job can outrun the seat price quickly.
Entry is unusually cheap — Starter at $3/mo and Creator at $7/mo put a small evaluation on-ramp within reach of a solo developer, though allowances are tiny (30,000 and 140,000 TTS characters). Serious throughput starts at Pro ($70/mo) and Scale ($200/mo with 3 seats); teams running continuous agent evaluation should budget Business ($500/mo) or Enterprise. Compared with all-in-one voice agent platforms, Hume is priced as a testing tool, not a deployment stack — you still pay separately for the
In short
Hume AI — Hume AI measures 600+ dimensions of vocal and facial expression to benchmark and improve voice AI agents. Best for Voice AI teams whose agents pass accuracy tests but feel wrong to real callers, Model developers comparing checkpoints and diagnosing weaknesses before release, Teams that want human-rated voice quality scores alongside automated LLM-judge results. Free to start; paid plans from $3/mo.
No head-to-head comparison for Hume AI yet — the alternatives below are the closest matches we list.
What is Hume AI?
Hume AI — Hume AI measures 600+ dimensions of vocal and facial expression to benchmark and improve voice AI agents. Best for Voice AI teams whose agents pass accuracy tests but feel wrong to real callers, Model developers comparing checkpoints and diagnosing weaknesses before release, Teams that want human-rated voice quality scores alongside automated LLM-judge results. Free to start; paid plans from $3/mo.
Hume AI built its name on measuring expression, and it now sells that measurement as voice AI evaluation infrastructure for the teams shipping speech agents and speech models. The core layer is Expression APIs, which score 600+ dimensions of vocal and facial signal across 50+ languages, catching tone, timing, and conversational dynamics that a transcript flattens. Audio and video endpoints return named expression scores per window, and realtime WebSocket endpoints plus Python and Node SDKs let you run it inside your own pipeline. On top of that sits an evaluation workflow for models and agents: define use-case-specific quality criteria around tone, timing, and turn-taking, stress-test across accents, noise, and interruptions, score with trained human raters, and compare checkpoints to see gains and regressions side by side. The Human Feedback API routes judgment to vetted raters, and the User Testing API recruits real participants against your agent endpoint across locales and survey dimensions like naturalness, ease, and trust. Hume also publishes public benchmarks. Real-World VoiceEQ ranks speech-to-speech and TTS systems on emotion alignment, naturalness, acting fit, and reliability, Voice Controllability tests whether a TTS model actually follows direction, and Voice Replication rates eleven models on reproducing a target voice. Newer to the lineup is the Audio Decisions API, an audio-native model that turns questions about audio into structured, probabilistic answers in one call instead of stitching together separate STT, speaker-ID, and language-ID models. This is for AI labs, voice-agent builders, and enterprises that need to know why conversations fail even when tasks succeed. It is not a voice-generation product: Hume sunset its TTS and EVI APIs on October 2, 2026, with access ending November 13, 2026. Against generic LLM-as-judge eval tooling or raw word-error benchmarks, Hume is the option purpose-built for the acoustic and emotional layer.
Key features
- Expression APIs score 600+ dimensions of expression across 50+ languages
- Audio endpoints return expression names and voice attributes per speech window
- Video endpoints return facial expression and description scores per detected face
- Realtime WebSocket endpoints for live audio and video measurement
- Python SDK and Node.js SDK with microphone and WAV helpers
- Audio Decisions API turns questions about audio into structured probabilistic answers
- VoiceEQ defines use-case-specific quality criteria for tone, timing, and turn-taking
- Stress-test voice agents across accents, noise, and interruptions
- Compare model checkpoints to surface gains, tradeoffs, and regressions
- Human Feedback API routes scoring to trained, vetted human raters
- User Testing API recruits real participants across locales and survey dimensions
- Real-World VoiceEQ leaderboard ranks speech-to-speech, TTS, and ASR systems
- Voice Controllability leaderboard tests whether TTS models follow directability
- Voice Replication leaderboard rates eleven TTS models on reproducing a target voice
- Benchmark baseline vs candidate checkpoints across TTS and speech systems
Behind the verdict
Most voice-agent evaluations grade the wrong thing. They check whether the intent was detected and the answer was right, then ship, and the caller hangs up feeling talked at. Hume's whole argument is that a task can succeed while the conversation fails, and its tooling is built to measure the gap: tone, timing, interruption handling, accent robustness, context, adaptation. If your roadmap already includes a voice quality bar beyond task completion, this is aimed straight at you. We'd reach for it in two scenarios. First, a model team comparing checkpoints before release — the workflow of defining criteria, running stress scenarios, then re-running to catch regressions is the right shape, and human raters on top of automated scores gives you a number you can defend internally. Second, an enterprise that needs third-party evidence of voice quality and can point to Hume's published benchmarks for context. The free leaderboards are the underrated part. Real-World VoiceEQ, Voice Controllability, and Voice Replication let you shortlist TTS and speech-to-speech models before you pay anything. Gemini 3.8 Flash TTS currently leads Real-World VoiceEQ, per Hume's September 2026 post. That's useful independent signal even if you never buy. Where it bites: this is measurement, not production. Hume does not run your calls or handle telephony, so you still need your own agent stack and an orchestration partner. Teams that wanted Octave TTS or EVI voices need a different vendor — those APIs sunset with access ending November 13, 2026 and account data deleted after. And if your product has zero emotional-nuance requirement, plain TTS plus word-error testing is cheaper and probably enough. Closest alternative is generic LLM-as-judge evaluation tooling. It's cheaper and easier to
Models under the hood
as of 2026-11-08
Limitations
- Hume's live site now centers on evaluation infrastructure and audio-native decision models rather than the older fixed TTS/EVI tiers, so underlying model versions cannot be verified from today's evidence.
- Measurement is offered through Expression APIs (upload, realtime WebSocket, run endpoints) with Python and Node.js SDKs, and the Audio Decisions API takes an audio file plus plain-language questions returning probabilistic answers.
- No pricing tiers are visible, and rate limits are documented only via changelog (600 req/min on Audio Decisions endpoints).
as of 2026-10-18
Researching Hume AI? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Is Hume AI still active?
Hume AI is live · free plan, paid from $3/mo · last verified 5 Nov 2026 · viability healthy.
Viability score
How well maintained and how widely used is Hume AI? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: November 2026
How we score →What's new in Hume AI
Checked 6 days agoAcross the latest 6 updates: 3 launches, 1 changelog entry and 2 news mentions.
Hume raises Audio Decisions API rate limit to 600 requests per minute
Per-org rate limit on Audio Decisions endpoints raised from 60 to 600 requests per minute, counted over a sliding 60-second window and shared across all API keys.
Introducing Hume's Audio Decisions API: a decision layer for speech and sound
Hume positions the API as a decision layer for call audio, extracting cues like pauses and overlap that transcripts miss so teams can score calls against a rubric without rebuilding pipelines.
Hume launches Audio Decisions API
Audio Decisions API lets developers send an audio file with plain-language questions and get back probabilities to threshold, rank or combine. Ships with Python and TypeScript SDKs.
Behind Hume's Expression Measurement models for face and voice
Hume published the science, development and evaluation behind its Expression API, which measures voice and facial expressions in real time.
Hume introduces Expression APIs for voice and facial expression measurement
Expression APIs give developers real-time measurement of vocal tone and facial expressions, aimed at adding emotional signal to applications and research pipelines.
Google's Gemini 3.8 Flash TTS tops Hume's Real-World VoiceEQ leaderboard
Hume reported that newly released Gemini 3.8 Flash TTS ranks first on its Real-World VoiceEQ benchmark, which Hume extended to cover voice controllability and voice replication.
What do users say about Hume AI?
What people actually say about Hume AI — is it worth it?
We scanned public community sources for Hume AI on Oct 12, 2026 and could not establish that the discussion we found is about this tool rather than something else sharing its name. Only 3 of the posts we fetched could be positively tied to Hume AI. Rather than publish a sentiment score built on the wrong subject, we publish nothing here and re-run the scan.
How we verified this
Every listed fact about Hume AI — pricing, features, links, status — is re-read from the vendor's own pages by our verification pipeline and each change is logged with a date; the last pass ran on 5 Nov 2026. Analysis is drafted with AI models and checked against our editorial standards before publishing; see how we verify.
Verification history
We have re-verified Hume AI 23 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — changed pricing page, limitations
- — changed summary, features, who it suits, who should skip it, description, our verdict, pricing tiers
- — changed who it suits, features, description, who should skip it, integrations, our verdict, summary, pricing tiers
- — changed limitations
- — changed our verdict, description, who it suits, pricing tiers, summary, features
- — changed link health
Showing the 6 most recent of 23 verification passes.
Free to cite with attribution — this page re-verifies continuously.
How much does Hume AI cost in 2026?
Hume AI: Free to start; paid plans from $3/mo. Pricing last verified 5 Nov 2026.
Plans compared
Every published Hume AI tier, its price as the vendor publishes it, who it fits, and what it adds over the tier below. The 12-month cost projection follows.
| Plan | Price | Billing | Who it's for |
|---|---|---|---|
| Free | Free | — | A developer who wants to see whether the Expression APIs fit before spending anything, or a student reproducing a leaderboard result.Adds: Free entry point: 10,000 TTS characters (~10 minutes), 5 EVI minutes, 15 requests per minute, 1 concurrent external-LLM connection, unlimited voice cloning create and use. |
| Starter | $3/mo | monthly | A solo builder prototyping a small voice feature who needs a bit more headroom than the free allowance.Adds: Adds 3x the TTS characters (30,000) and 8x the EVI minutes (40) over Free, and raises concurrent external-LLM connections from 1 to 5; RPM stays at 15. |
| Creator | $14/mo (intro $7)Vendor: “$7/mo first month, $14/mo after” | monthly | An indie developer or small studio running regular short evaluations and light synthesis.Adds: Jumps to 140,000 TTS characters and 200 EVI minutes and lifts the rate limit from 15 to 75 requests per minute, with additional TTS at $0.15 per 1,000 characters. |
| Pro | $70/mo | monthly | A product team running continuous agent evaluation across a real user base.Adds: 1,000,000 TTS characters and 1,200 EVI minutes, 10 concurrent external-LLM connections, and a lower additional-EVI rate of $0.06/minute. |
| Scale | $200/mo | monthly | A growing voice team that needs volume plus shared access for a few engineers.Adds: 3,300,000 TTS characters, 5,000 EVI minutes, 150 requests per minute, 20 concurrent connections, and 3 team seats included. |
| Business | $500/mo | monthly | An established company running voice evaluation in production across multiple squads.Adds: 10,000,000 TTS characters, 12,500 EVI minutes, 225 requests per minute, 30 concurrent connections, 5 team seats, and Slack support. |
| Enterprise | Custom | quote | Regulated buyers in healthcare, insurance, or finance that need compliance documentation and custom throughput.Adds: Unlimited TTS and EVI usage with custom RPM and overage pricing, API access for voice cloning, and SOC 2 Type II, GDPR, and HIPAA compliance. |
12-month cost
Project the real annual outlay, including the implied monthly cost when only an annual tier is published.
Vendor list price only. Add-on usage, seat overages, and contract minimums are surfaced under Hidden costs & gotchas.
Where the pricing makes sense
The company stage and team size where Hume AI's pricing actually pencils out — and where peers do it cheaper.
Entry is unusually cheap — Starter at $3/mo and Creator at $7/mo put a small evaluation on-ramp within reach of a solo developer, though allowances are tiny (30,000 and 140,000 TTS characters). Serious throughput starts at Pro ($70/mo) and Scale ($200/mo with 3 seats); teams running continuous agent evaluation should budget Business ($500/mo) or Enterprise. Compared with all-in-one voice agent platforms, Hume is priced as a testing tool, not a deployment stack — you still pay separately for the
Has Hume AI's pricing changed?
No plan price for Hume AI has moved since we began logging in Aug 2026; the pricing page is re-read on every verification pass.
Who should pick Hume AI?
Pick Hume AI if you are Voice AI teams whose agents pass accuracy tests but feel wrong to real callers, Model developers comparing checkpoints and diagnosing weaknesses before release, Teams that want human-rated voice quality scores alongside automated LLM-judge results, Companies running private evaluations of TTS, speech-to-speech, or ASR systems.
Look elsewhere if you are Teams shopping for voice generation — TTS and EVI access ends November 13, 2026, Buyers wanting a turnkey voice agent platform with telephony and call handling, Developers who need a fully open-source voice stack they can self-host and fork, Projects with no emotional-nuance requirement, where basic TTS or STT is cheaper.
Skip it if Skip Hume if you need a voice model to generate with rather than a way to measure one — the TTS and EVI APIs are sunsetting November 13, 2026, and the value here is evaluation, not synthesis.
GeneratedReal-world workflow fit
Concrete scenarios for the personas Hume AI actually fits — and what changes day-one when you adopt it.
Your agent resolves tickets but CSAT is slipping. You feed real support transcripts into VoiceEQ, which generates a criteria list, metric set, and experiment plan, then replay agent-to-agent scenarios to find where barge-ins and flat tone break the conversation.
Outcome: You get a ranked list of failure moments by dimension (timing, emotion, accent, adaptation) instead of a single accuracy score, plus a regression suite you can rerun after every prompt or model change.
You have a baseline and a candidate voice model. You run both through the Expression APIs and compare on the Real-World VoiceEQ dimensions — expression, naturalness, timing and turn-taking, contextual appropriateness, transcript accuracy — then check the Voice Controllability leaderboard to see whether your model follows direction at all.
Outcome: You ship or hold the checkpoint on evidence, and you know which dimensions moved (baseline 0.62 to candidate 0.81 on TTS in Hume's illustrative example) rather than guessing from listening tests.
Your automated scores disagree with what your team hears. You route the disputed samples through the Human Feedback API to trained, vetted raters and compare per-vertical results — Hume's own sample shows healthcare at 0.84 automated versus 0.76 human.
Outcome: You learn where your automated evaluator is wrong before it steers a release, and you have per-sample human ratings to defend the call.
Use cases
- Measure whether a contact-center voice agent reduces caller frustration using the Expression Measurement API
- Run Human Feedback API studies to score TTS and speech-to-speech models against trained human raters
- Turn real support transcripts into VoiceEQ evaluation suites and replay them as regression tests
- Check which TTS model actually follows your direction on the Voice Controllability leaderboard
- Compare a candidate TTS checkpoint against a baseline before shipping a voice model update
- Pressure-test a healthcare or insurance voice agent with agent-to-agent conversations before real callers hit it
- Add real-time vocal expression and turn-taking signals to a speech-to-speech assistant
- Benchmark ASR and cascaded systems on transcript accuracy plus the qualities transcripts miss
Setup time & first value
How long it actually takes to get something useful out of Hume AI — broken out by persona, not the marketing-page minute.
API access is the gating step: the Expression Measurement API is in limited release and requires requesting access, so plan on a sales/access conversation before your first call. Once you have a key, the Python or Node.js SDK plus the audio helpers (microphone capture, WAV reading, format conversion) get a first measurement running in an afternoon. VoiceEQ suite generation from existing
Switching to or from Hume AI
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From manual listening tests: upload the same recorded calls to the Expression API and get per-window scores for 414 expression names instead of reviewer notes.
- →From a generic TTS benchmark: run your own transcripts through VoiceEQ to build a use-case-specific suite rather than relying on a public leaderboard score.
- →From Octave TTS or EVI generation: note access ends November 13, 2026, so migrate synthesis to another vendor and keep Hume for measurement only.
- →From an LLM-judge-only eval pipeline: add the Human Feedback API to calibrate where automated scores diverge from trained raters.
- ↗To ElevenLabs, Cartesia, or Gemini 3.8 Flash TTS: move voice generation there — Hume's leaderboards rank all three — and keep Hume for evaluation if you still need measurement.
- ↗To a turnkey voice agent platform: if you need telephony and call handling in one box, Hume was never that layer; export your evaluation criteria and rebuild them in the platform's own test harness.
- ↗To an open-source voice stack: Hume's APIs are hosted and the Expression Measurement API is limited-release, so self-hosting means replacing the measurement layer entirely.
Integrations
Resources & guides
- Documentationhume.ai
Welcome to Hume AI | Hume API
Hume AI builds AI models that enable technology to communicate with empathy and support human well-being.
- Documentationhume.ai
Support | Hume API
Get technical support, contact our team, or explore enterprise and research programs.
- Resourcehume.ai
The AI toolkit for voice and emotion
Building AI with emotional intelligence to create technology that truly understands humanity.
Tutorials & Learning
YouTube returned 6 videos for “Hume AI”, and we withheld 5: 5 could not be judged, because “Hume AI” is a single word that other videos use for other things. Showing the 1 we can prove is about Hume AI.
Official links
What are the best Hume AI alternatives?
The closest alternatives to Hume AI we list are ElevenLabs, Cartesia and Murf AI. Ranked by direct product-type match, not category overlap.
| Alternative | Pricing | Starts at | Last verified |
|---|---|---|---|
| ElevenLabs | Freemium | Pricing not published | — |
| Cartesia | Freemium | Pricing not published | — |
| Murf AI | Freemium | Pricing not published | — |
Alternatives to Hume AI
View allElevenLabs
AI voice platform for text to speech, cloning, dubbing and conversational agents — now with music, SFX and generated video.
Tools that pair well with Hume AI
Common stack mates teams adopt alongside Hume AI, with the specific reason each pairing earns its keep.
ElevenLabs
AI voice platform for text to speech, cloning, dubbing and conversational agents — now with music, SFX and generated video.
Cartesia
Cartesia ships real-time text-to-speech, streaming speech-to-text and Managed Agents for live voice agents
Murf AI
Murf AI turns scripts into voiceovers and text-to-speech into live conversational voice agents.
Frequently asked questions
Best-of guides
For makers of Hume AI
- Get a Verified Listing for Hume AI ($39): a written decision within 1 working day.
- Free listing for another tool: Free reviews are decided in order, within 3 weeks of the review starting.
Used Hume AI? Help shape our editorial sentiment research.