ElevenLabs v4 Review: What's Actually New (2026)
We may earn a commission if you buy through links on this page. Read the affiliate disclosure · How we review tools
Quick answer. We review ElevenLabs v4: what's new vs v3, vendor claims vs independent tests, real limitations, and when Murf, Fish Audio & others win.
| Tool | Rating | Best for | Price | Website |
|---|---|---|---|---|
| 4.6/5 | Best-sounding voice model overall | Free plan; paid from $6/mo | Visit ElevenLabs v4 → | |
| 4.5/5 | Video & course voiceovers | Free plan; paid from $29/mo | Visit Murf AI → | |
| 4.6/5 | Expressive TTS & voice cloning | Free plan; paid from $11/mo | Visit Fish Audio → | |
| 4.3/5 | Reading & listening to text | Free plan; Premium ~$139/yr | Visit Speechify → | |
| 4.2/5 | Developers & rapid cloning | Pay as you go from $0 | Visit Resemble AI → | |
| 4.3/5 | Enterprise & compliance teams | Paid from ~$50/user/mo | Visit WellSaid Labs → | |
| 3.9/5 | Budget creators | Trial; paid from $11/mo | Visit DupDub → |
ElevenLabs launched v4 on September 28, 2026, and called it the new #1 voice model: most expressive ever, 90+ languages, 10-second voice cloning, ~100ms latency on v4 Turbo. Some of that checks out against independent data. Some is vendor-reported. And a few details — a disabled SSML tag, a consent question the launch never addressed, clones that need retraining — only surface when you read past the announcement.
How we verify: we did not run the ElevenLabs API or the competitors’ apps for this review. Every claim below was cross-checked between ElevenLabs’ launch materials and API docs, dated third-party coverage from September–October 2026, and independent leaderboards — chiefly Artificial Analysis’ Speech Arena. Vendor-sourced figures are labeled as such. Prices were checked against official pricing pages and reputable October 2026 roundups; plans change, so confirm on the vendor’s site before buying.
Short answer: ElevenLabs v4 is a real generational step. The context-aware delivery model, 90+ language coverage, 10-second instant cloning, and the return of Professional Voice Clones make it the best-sounding voice model you can buy — and the independent Artificial Analysis Speech Arena agrees: as of mid-October 2026 the v4 family holds the top two spots, v4 Turbo at ~1334 Elo just ahead of full v4 at ~1321, both well clear of Cartesia’s Sonic 3.6 at ~1278. (Elo moves as votes accumulate — the board’s current numbers, not September’s, are what counts.) Honest caveats: SSML break tags are disabled, the ~100ms latency figure is vendor-reported, the API list price is steep next to Google’s Gemini 3.8 Flash TTS, and old voice clones must be retrained. Upgrade if you work in Japanese, Brazilian Portuguese, Mandarin, or Cantonese, or need better long-form consistency. If v4’s pricing or limitations bother you, Fish Audio is the quality challenger, Murf AI is the workflow pick, and DupDub is the budget option.
Quick Comparison
| Tool | Starting price | Voice quality | Voice cloning | Languages | Free plan |
|---|---|---|---|---|---|
| ElevenLabs v4 | $6/mo (30k chars) | #1 on Speech Arena | Yes, 10-sec instant | 90+ | Yes, 10k chars/mo |
| Murf AI | $29/mo ($19 annual) | Excellent | Enterprise only | 35+ | Yes, 10 min lifetime |
| Fish Audio | $11/mo | Excellent | Yes, ~10-sec sample | 80+ | Yes, free tier |
| Speechify | Free / ~$139/yr Premium | Very good (reading) | No | 50+ | Yes, limited |
| Resemble AI | $0 (pay per use) | Very good | Yes, rapid clone | 100+ | Test credits |
| WellSaid Labs | ~$50/user/mo | Excellent | Enterprise only | English-only | 7-day trial |
| DupDub | $11/mo | Good | Yes | 90+ | 3-day trial |
Pricing checked October 2026 — confirm live on the official page before buying.
What’s Actually New in v4
Five genuine changes:
1. A context-aware delivery model. v3 read sentences; v4 reads scenes. The new architecture tracks who is speaking, what happened earlier in the passage, and how each line should land, then adjusts tone and pacing across the whole script. Multi-speaker dialogue now responds to what was just said instead of rendering isolated lines. This matters most for audiobooks, dubbing, and long-form work — and it is architectural, not cosmetic, so it is the hardest thing for a competitor to replicate.
2. 90+ languages, up from roughly 70. ElevenLabs says the biggest quality gains are in Japanese, Brazilian Portuguese, Mandarin, and Cantonese, with coverage extending to languages like Mongolian and Odia, plus 14 Indian languages announced. Independent verification of the smaller languages is thin.
3. 10-second instant cloning. A usable voice clone from ten seconds of reference audio, down from the longer samples v3 needed. Cross-language cloning keeps the speaker’s identity while speaking the target language with a natural local accent. Two honest footnotes: older Instant and Professional Voice Clones need retraining to work with v4, and the target-language accent is currently applied automatically — retaining the original accent is “being explored” with no timeline.
4. Stackable audio tags replace SSML pauses. v3 introduced inline tags; v4 lets you stack them — [laughs] followed by [said angrily in French accent] — with the model following the sequence and using surrounding context to cut manual annotation. The price: SSML <break> tags are disabled. If your pipeline is built on precise SSML markup, v4 forces a rewrite of your pause logic. The single most annoying migration cost in v4.
5. Professional Voice Clones are back. PVC, the highest-fidelity cloning tier, returns after being unavailable in v3. All 17,500+ library voices work with v4, though existing clones need retraining.
v4 Turbo, the twin release, is the speed play: ElevenLabs reports roughly 100ms median latency, tuned for conversational agents. That figure is vendor-reported from the company’s own testing — and a reviewer at TBreak made the necessary point that total response time in a real product also includes the language model, the network, and the application layer. The practical consequence arrived October 7: ElevenAgents’ default voice model switched to v4 Turbo, so agents that never pinned a model silently changed behavior — check your output if that is you.
v4 vs v3: Should You Upgrade?
| Eleven v3 | Eleven v4 | |
|---|---|---|
| Languages | ~70 | 90+ |
| Instant cloning sample | Longer audio | 10 seconds |
| Emotion/delivery tags | Basic inline tags | Stackable, context-aware |
| Professional Voice Clones | Unavailable | Supported |
| SSML break tags | Supported | Disabled |
| Long-form consistency | Decent | Context stitching across passages |
| 17,500+ library voices | Supported | Supported, after retraining older clones |
| Median latency (Turbo twin) | — | ~100ms (vendor-reported) |
Upgrade now if: you produce in Japanese, Brazilian Portuguese, Mandarin, or Cantonese; you do long-form narration, dubbing, or multi-speaker dialogue; you want the cheapest path to a high-fidelity clone; or you run voice agents and want the Turbo latency profile.
Wait if: your workflow is English-only narration with pinned v3 settings; your pipeline depends on SSML <break> tags; or you have a large library of trained clones you would need to rebuild.
ElevenLabs v4: The Honest Scorecard
👍 Pros
- v4 family ranked #1 and #2 on the independent Artificial Analysis Speech Arena (v4 Turbo ~1334 Elo, checked mid-October 2026)
- Context-aware delivery: multi-speaker dialogue and long-form consistency that v3 could not do
- 10-second instant cloning with cross-language identity transfer
- Professional Voice Clones return; 90+ languages; ElevenReader and ElevenAgents already moved to v4
👎 Cons
- SSML break tags are disabled — existing pipelines need rework
- Key latency figures are vendor-reported, never independently measured
- API list pricing is steep next to Google and Cartesia; the 10-second cloning launch skipped the consent discussion
Vendor Claims vs Independent Data
Claim: “Ranked #1.” This one holds with a nuance the launch materials gloss over. Artificial Analysis’ Speech Arena — an independent, blind-vote leaderboard, not an ElevenLabs property — ranks the v4 family first and second as of mid-October 2026: v4 Turbo at ~1334 Elo just ahead of full v4 at ~1321, both well clear of Alibaba’s Qwen-Audio-3.1-TTS-Plus at ~1295, Cartesia’s Sonic 3.6 at ~1278, and Google’s Gemini 3.8 Flash TTS at ~1276. The board is a living vote count — a read on October 8 showed v4 Turbo at 1332 and v4 at 1324 — so treat “v4 is #1” as true for the family, not strictly for v4 itself, and attach the date whenever you quote a score. ElevenLabs separately cites ~75% preference in its own head-to-head tests — the company’s own measurement, not a third party’s. Trust the Arena, discount the rest.
Claim: “~100ms median latency.” Vendor-reported. Two independent outlets repeated the figure from the launch materials — widely cited, still originating with ElevenLabs. Nobody outside the company has published independent latency measurements at this writing. The vibeleaderboard page tracking v4 Turbo carries an explicit caveat: its figures are vendor-reported, and it recommends running your own scripts to verify. If latency is your purchase criterion, take that advice.
Claim: “best cloning.” This is where honesty requires a split decision. On the Controlled Voice Arena, which isolates pure cloning fidelity, results are more scattered than ElevenLabs’ marketing suggests: Alibaba’s Qwen-Audio-3.1-TTS-Plus currently leads that narrow test. ElevenLabs v4 wins the general Speech Arena crown, but “best voice model” and “best clone” are different questions, and the independent data does not give ElevenLabs both titles. Meanwhile Fish Audio’s Drama 3 (October 13, 2026) scored 93.2% on Artificial Analysis’ Pronunciation Robustness benchmark — the top score on that board. ElevenLabs’ pronunciation lead is now contested by an open-weight model.
Claim: 90+ languages. Partially verifiable. The headline count is ElevenLabs’ own. The specific claim about Japanese, Brazilian Portuguese, Mandarin, and Cantonese gains was corroborated by TechCrunch’s launch coverage.
The consent question the launch never addressed. VoiceOverHerald’s October 6 coverage made a fair point: ElevenLabs’ launch materials for 10-second cloning said nothing about consent — while Google requires 30 seconds of audio plus an explicit voice confirmation for its cloning-equivalent features. ElevenLabs does have verification requirements on its Professional tier, but the flagship headline shipped with no consent discussion in the announcement. A genuine reason some buyers prefer vendors like WellSaid Labs, whose voices are built on licensed, consent-based voice-actor data.
Pricing: What v4 Actually Costs
Subscription pricing did not change with v4. Checked October 2026:
- Free: $0 — 10,000 characters per month, personal use only
- Starter: $6/mo — 30,000 characters
- Creator: $22/mo — 121,000 characters (first-month half-price, $11, still active at last check)
- Pro: $99/mo — 600,000 characters
- Scale: $299/mo — 1.8 million characters
- Business: $990/mo — 6 million characters
Annual billing runs roughly 10x the monthly price.
For developers, the API story matters more. List pricing is $0.08 per 1,000 characters for v4 and $0.04 for v4 Turbo — $80 and $40 per million characters. The 72%-off launch discount (down to $0.022 and $0.011) ended around October 12, 2026, so do not budget on it. At list price, v4 costs $80 per million characters against $49 for Cartesia’s Sonic 3.6 and $16.50 for Gemini 3.8 Flash TTS on the same Artificial Analysis comparison table.
Two ecosystem notes from October: ElevenReader switched to v4 on October 10, so articles, ebooks, and PDFs now read aloud in the new model even on the free tier. And ElevenAgents moved its default voice model to v4 Turbo on October 7 — if you never pinned a model, your agent’s voice changed without you touching anything.
Try ElevenLabs →The Alternatives: How They Compare to v4
1. Murf AI: Best for Video & Course Voiceovers
Murf AI does not try to out-sound ElevenLabs. It wins the workflow: a studio editor where script, voices, timing, music, and slides live on one timeline. For explainer videos, courses, and corporate training, that saves more hours than any voice-quality edge ElevenLabs holds.
The library runs 200+ voices across 35+ languages, with PowerPoint and Google Slides plugins. Voice quality is excellent for narration and clearly weaker than ElevenLabs on drama and character work. The structural gap is cloning: a managed, Enterprise-only service — up to 90 minutes of studio recordings and one to four weeks with Murf’s team — while instant cloning is API-only. Neither self-serve plan includes cloning.
Pricing, checked October 2026: free plan with 10 minutes of lifetime generation (no downloads, no commercial rights), Creator at $29/mo or $19/mo billed annually, Business at $99/mo or $66/mo annually.
👍 Pros
- Best-in-class studio workflow: script, voice, music, timing, and slides on one timeline
- Polished voices ideal for corporate narration, e-learning, and explainers
- Segment regeneration saves real editing time
- Commercial rights on all paid plans
👎 Cons
- Voice cloning is Enterprise-only — a hard gap against ElevenLabs' 10-second clones
- Emotional and dramatic range lags ElevenLabs v4
- Time-based allowances burn fast if you regenerate often
Best for: Course creators, YouTubers, and L&D teams who want script-to-finished-video in one place.
Why people switch from ElevenLabs to Murf: one workspace from script to finished video, no separate editor.
Try Murf AI →2. Fish Audio: Best Quality Challenger to v4
Fish Audio is the one competitor ElevenLabs should be watching. Born from the open-source TTS community — the team behind So-VITS-SVC and Bert-VITS2 — it has a pattern of showing up near the top of independent leaderboards. Its S1 model topped the TTS-Arena2 blind-test leaderboard, and the new Drama 3, released October 13, 2026, leads Artificial Analysis’ Pronunciation Robustness benchmark at 93.2%. The weights are open and the Community License is free for commercial use under $500,000 in annual revenue.
With Drama 3 you direct the performance with plain-language notes in square brackets, mid-line, in 83 languages — emotion, breath, pauses — more like directing an actor than filling a text box. Cloning works from roughly a 10-second sample, and the library is a 2,000,000+ voice community ecosystem. The honest caveats: no Murf-style studio workflow, no enterprise compliance packaging, and post-launch API pricing is still settling.
Pricing, checked October 2026: free tier with monthly credits, Plus from $11/mo with commercial rights, Pro at $75/mo for heavy use, API pay-as-you-go with sub-500ms streaming latency.
👍 Pros
- Independent leaderboard results that genuinely challenge v4 — S1 topped TTS-Arena2, Drama 3 leads pronunciation benchmarks
- Directable, actor-style performance control unique at this price
- Open weights with a permissive community license for small businesses
- Cloning from ~10 seconds, aggressive API pricing for developers
👎 Cons
- No built-in studio workflow for assembling productions
- Community voice library varies in quality
- No enterprise compliance story — regulated buyers look elsewhere
- Post-launch API pricing is still settling
Best for: Creators who want the best-sounding voice short of ElevenLabs, and developers who want open weights or cheap real-time voice.
Why people switch from ElevenLabs to Fish Audio: comparable-or-better benchmark scores at a lower price, plus open weights.
Try Fish Audio →3. Speechify: Best for Reading & Listening
Speechify plays a different game. It began as a read-it-later app and remains strongest at turning documents, PDFs, web pages, and emails into listenable audio — including accessibility use cases, where its OCR reader and large, highly rated user base matter. It has been expanding its partnerships through 2026.
As an ElevenLabs alternative, the honest assessment is narrower. Speechify is the best pick if your job is consuming text as audio — studying, reviewing documents, listening on a commute. It is not the first pick for producing voiceovers: narration-studio tooling, emotion control, and cloning do not match the dedicated voice platforms. There is no meaningful voice cloning, and the catalog is built for pleasant long-form listening, not character performance.
Pricing, checked October 2026: limited free tier, Premium around $139/year — cheaper than ElevenLabs’ Creator tier for a listener, not comparable for a producer.
👍 Pros
- Best-in-class reading workflow: documents, PDFs, OCR, cross-device sync
- Strong accessibility story and a genuinely large, happy user base
- Cheaper than voice-production tools if listening is the job
- Actively growing — major partnerships and new investors in 2026
👎 Cons
- Not a voiceover production tool — weak emotion and delivery control
- No meaningful voice cloning
- Voice catalog tuned for listening, not performance
Best for: Students, researchers, professionals, and anyone with accessibility needs who wants their reading read aloud.
Why people switch from ElevenLabs to Speechify: their real job is listening to text, not producing voiceovers.
Try Speechify →4. Resemble AI: Best for Developers
Resemble AI has repositioned itself in 2026, and buyers should know. The headline business is now AI security — deepfake detection (its DETECT-3B Omni detection model), watermarking, and identity verification — backed by fresh strategic funding in late 2025. Voice generation and cloning remain as one pillar of a “generate, verify, detect” platform. Healthy and funded — just no longer a pure TTS vendor.
For voice work, Resemble remains cloning-first: rapid clones from around 5 seconds of reference audio, 100 languages and dialects claimed, and the MIT-licensed open-source Chatterbox TTS family for self-hosting. Pricing is Flex pay-as-you-go — start at $0, roughly $0.0005 per second of TTS output — plus custom Enterprise contracts. Good for bursty workloads, poor for always-on production traffic.
The tradeoff is focus. The web studio is secondary to the API, and roadmap energy is visibly pointed at detection. If voice authenticity is a legal concern, the generate-plus-detect combination is unique. For the best-sounding narration, v4 or Fish Audio win.
👍 Pros
- Fastest self-serve cloning on this list — around 5 seconds of audio
- Pay-per-use Flex pricing with no wasted subscription months
- Deepfake detection and watermarking built in — unique combination
- MIT-licensed open-source Chatterbox models for self-hosting
👎 Cons
- Company focus is shifting toward security, not consumer voice quality
- Web studio weaker than Murf or ElevenLabs for pure content creation
- Pricing favors bursty use; heavy always-on traffic needs Enterprise quotes
Best for: Developers building voice features into apps, and teams that need generation and deepfake detection from one vendor.
Why people switch from ElevenLabs to Resemble: cheaper cloning at low volume, plus detection tooling for compliance.
Try Resemble AI →5. WellSaid Labs: Best for Enterprise & Compliance Teams
WellSaid Labs is the vendor your legal team approves fastest. Voices are built through consent-based partnerships with professional voice actors, outputs carry guaranteed commercial rights, and the company holds SOC 2, GDPR, and HIPAA compliance. Clients include T-Mobile, ServiceNow, and Accenture.
Voice quality is genuinely excellent for professional narration, and AI Director gives word-level control over emphasis, pacing, and pronunciation — what training-video producers need when a brand name must be said identically every time. The tradeoffs are structural: English-only voices, no self-serve voice cloning (custom voices are an Enterprise conversation), and per-user pricing from around $50/user/mo on standard tiers. No permanent free plan — only a trial. There is an enterprise Partner Program for ISVs, but no public publisher affiliate program, which tells you who the product is for.
👍 Pros
- Broadcast-quality output with licensed, ethically sourced voices
- Best-in-class compliance packaging (SOC 2, GDPR, HIPAA) for regulated industries
- AI Director gives precise, repeatable control over delivery
- Financially stable enterprise business with blue-chip clients
👎 Cons
- English-only voices — a hard limit for multilingual teams
- Per-user pricing gets expensive fast for larger teams
- No self-serve cloning; no permanent free plan
Best for: Enterprise L&D, healthcare, finance, and any team whose compliance department has veto power.
Why people switch from ElevenLabs to WellSaid: procurement requirements and licensed voice data that survives legal review.
Try WellSaid Labs →6. DupDub: Best Budget Pick
DupDub is the value play, and the pricing is refreshingly simple: a 3-day free trial, then Personal at $11/mo, Professional at $30/mo, and Ultimate at $110/mo — consistent across multiple 2026 sources, with commercial rights on paid plans. For that you get 500+ voices across 90+ languages and accents, voice cloning, AI avatars, talking photos, and video translation. Nothing else here bundles this much for $11.
The honest tradeoff is voice quality: a half-step behind the top tier — flatter intonation, occasional odd pacing on long sentences, less convincing emotional delivery. For short-form content, social videos, and faceless channels, it is more than good enough, and the avatar and translation features are things ElevenLabs does not offer at this price. The credit-based billing takes some budgeting to understand, and the community is smaller than the big platforms’.
👍 Pros
- Lowest entry price here with real features — $11/mo with cloning and commercial rights
- Avatars, talking photos, and video translation bundled in
- 90+ languages with voice cloning included
- Pays for itself after one or two voiceover projects versus hiring talent
👎 Cons
- Voice quality a notch below ElevenLabs v4, Murf, and Fish Audio
- Credit-based billing takes effort to budget
- Smaller community and fewer third-party tutorials
Best for: Budget-conscious creators, faceless channels, and anyone testing whether AI voiceover fits their workflow.
Why people switch from ElevenLabs to DupDub: avatars plus voiceover in one $11 subscription, without needing top-tier realism.
Try DupDub →The Competitive Backdrop
Around the release: Google shipped Gemini 3.8 Flash TTS on September 23 with a cheaper Flash-Lite tier — at $16.50 per million characters it undercuts v4’s $80/M list price by nearly 5x. Suno launched Speech on October 2, entering voice from the music side. And OpenAI’s GPT-4o-mini-TTS became the default voice option in its developer stack, quietly normalizing “good enough” voice inside the biggest developer ecosystem on earth.
None of these dethrone v4 on quality — the Speech Arena ranking stands. But they compress the pricing umbrella ElevenLabs sits under. The moat is quality; the pressure is price.
One more signal on the company’s staying power, which matters when you’re picking a vendor to build on: on September 30, 2026, ElevenLabs completed a $300 million employee tender offer at a $22 billion valuation — double its $11 billion Series D valuation from February 2026. The tender was led by Wellington and T. Rowe Price, with participation from Goldman Sachs and Singapore’s GIC, first reported by the Financial Times and independently confirmed by Reuters the same day. That’s institutional money buying into the company’s existing momentum, not a rescue round.
Buyer’s Guide: Pick by Your Job
- Maximum realism, audiobooks, dubbing: ElevenLabs v4. The Arena data backs it.
- Video and course production: Murf AI. Script-to-finished-voiceover in one studio.
- Best quality per dollar / open weights: Fish Audio. The challenger with independent results.
- Listening to text, studying, accessibility: Speechify. A different job, done better and cheaper.
- Building a voice app or needing detection: Resemble AI. Pay-per-use API plus deepfake tooling.
- Enterprise with compliance requirements: WellSaid Labs. Licensed voices that pass legal.
- Tightest budget: DupDub. $11/mo with cloning, avatars, and translation.
- Real-time voice agents: ElevenLabs v4 Turbo — with the caveat that the latency figures are vendor-reported.
Final Verdict
ElevenLabs v4 earns its #1 ranking on the independent Speech Arena, and the upgrade is genuinely worthwhile for multilingual creators, long-form narrators, and anyone cloning voices — 10 seconds of audio for a usable clone is a real threshold crossed. The honest ledger: the launch discount is over, the API list price is steep against Google’s, SSML break tags are gone, old clones need retraining, the latency numbers are vendor-reported, and the launch said nothing about consent for 10-second cloning. Fish Audio’s Drama 3 proves the open-weight challengers are not standing still.
If voice quality is the job, v4 is the best tool you can buy — prove it on the free tier first. If workflow, budget, compliance, or open weights matter more, the alternatives above are honest picks, not consolation prizes.
Try ElevenLabs →Pricing last verified October 2026. This article contains affiliate links: we may earn a commission if you purchase through them, at no extra cost to you. We did not run hands-on tests for this review — all performance claims were cross-checked between vendor documentation, dated third-party coverage, and independent leaderboards, with vendor-reported figures labeled as such.
Frequently asked questions
Is ElevenLabs v4 worth the upgrade from v3?
For most users, yes — especially in Japanese, Brazilian Portuguese, Mandarin, and Cantonese, where ElevenLabs says the biggest gains are. Instant cloning drops to a 10-second sample, emotion tags stack more reliably, and Professional Voice Clones return. If your workflow is English-only narration with pinned v3 settings, you can wait.
What is the latency of ElevenLabs v4 Turbo?
ElevenLabs reports a median latency of about 100ms for v4 Turbo. That is a vendor-reported figure from the company's own testing; real-world latency also depends on the language model, the network, and the application layer.
Does ElevenLabs v4 support SSML?
Only partially. SSML break tags are disabled in v4 — pauses and delivery are controlled through natural-language audio tags like [pauses] or [whispers] instead. If your pipeline depends on precise SSML markup, this is a genuine migration cost.
Can I clone my voice with ElevenLabs v4?
Yes. Instant cloning works from a 10-second sample, and Professional Voice Clones — the highest-fidelity option — are supported again after being unavailable in v3. Note that older Instant and Professional clones need retraining to work with v4.
How much does ElevenLabs cost after the v4 launch?
Subscription tiers are unchanged: Free ($0), Starter ($6/mo), Creator ($22/mo), Pro ($99/mo), Scale ($299/mo), Business ($990/mo), with annual billing at roughly 10x the monthly price. API list pricing is $0.08 per 1,000 characters for v4 and $0.04 for v4 Turbo; the 72%-off launch discount ended October 12, 2026.
Which alternative comes closest to ElevenLabs v4 quality?
On independent leaderboards, Fish Audio is the closest challenger — its models have topped blind-test rankings, and the new Drama 3 (October 2026) leads Artificial Analysis' Pronunciation Robustness benchmark at 93.2%. For production workflows, Murf AI is the closest all-round alternative; for enterprises, WellSaid Labs.