Short Answer

Amazon is not expected to be the top Text to Speech AI this week, given Amazon Polly Generative's #59 ranking on the Artificial Analysis Speech Arena leaderboard. Its recent upgrade focused on production and catalog expansion, not material performance improvement, with the market pricing this outcome at 3.0%.

1. Market Behavior & Drivers

The primary dynamic in this market was a sharp decline in the contract for ElevenLabs on October 26, 2026, which saw its implied probability fall by 87 percentage points. No reported development or specific news event coincides with this repricing. The move reflects a major re-evaluation of the competitive landscape by market participants, detached from any clear external catalyst.
Following this event, the market structure shifted from anticipating a single frontrunner to pricing a wide-open field. A cluster of competitors—including Cartesia, Google, Smallest.ai, MiniMax, and Inworld—now carry similar implied probabilities between 10% and 11%. ElevenLabs is currently priced at 1%, indicating traders have effectively removed it from contention for the top position on the Artificial Analysis Speech Arena leaderboard by the market's resolution date.
  • Cartesia, Inworld, Smallest.ai, MiniMax, and Mistral lead, maintaining existing market positions.
  • ElevenLabs' high Elo ratings face market volatility for the leaderboard resolution.
  • Amazon Polly's low ranking persists; its update focused on catalog expansion, not performance.

Who Wins and Why

Outcome Market Model Why
ElevenLabs 1.0% 1.8% ElevenLabs v4/v4 Turbo models consistently rank at or near #1 in Artificial Analysis Provider Voice Speech Arena.
Amazon 3.0% 2.0% Amazon Polly Generative ranks #59 on Artificial Analysis, making a #1 position unlikely despite a recent update.
Google 10.0% 2.0% Outside deterministic model scope.
Alibaba 9.0% 2.0% Outside deterministic model scope.
Nvidia 8.0% 2.0% Outside deterministic model scope.

Current Context

Prediction market shows varied probabilities for top TTS AI. The Oct. 30, 2026 prediction market for top Text-to-Speech (TTS) AI, resolving at 2:00 PM UTC based on the Artificial Analysis Speech Arena leaderboard, indicates diverse implied probabilities for leading contracts [1]. A public snapshot taken between Oct. 23-30, 2026, showed Cartesia at 11%, Google at 10%, Alibaba at 9%, Nvidia at 8%, Smallest.ai at 10%, MiniMax at 10%, Inworld at 10%, and ElevenLabs at 1% [1][2]. These figures represent market prices, not objective performance probabilities [1]. Kalshi's indexed page did not provide reliable live odds, making the Public snapshot the usable odds source [1][2].
ElevenLabs leads conventional TTS quality, others serve specific niches. For conventional text-to-speech quality, ElevenLabs v4 and v4 Turbo demonstrate leading performance as of late October 2026 [3][4][5][6]. One comparison reported v4 Turbo at 1,334 Elo and v4 at 1,321 Elo [4]. Practical selection often depends on use case: ElevenLabs is recommended for premium narration, voice cloning, and expressive customer-facing voices [5][6]. Google Cloud and Amazon Polly offer advantages for cost efficiency, enterprise integration, and broad language support [5][6]. Cartesia or ElevenLabs Flash are noted for low-latency streaming applications [5][6]. Azure provides strong options for governance, branded voice controls, and auditability [5][6]. Amazon Polly enhanced its Generative engine on October 22, 2026, adding 14 voices across nine locales, including generally available Hindi and Finnish generative voices [7]. Looking ahead, Gemini 3.8 Flash TTS promotional prices are scheduled to double on January 1, 2027, which could impact its current cost advantages [3][5].
Voice AI innovation focuses on real-time, full-duplex agents. The week of October 19–25, 2026, saw significant developments beyond traditional TTS, with a frontier shift towards real-time, full-duplex voice agents [8][9][10]. Mistral launched Voxtral 3, which includes Mini TTS 3 and supports over 35 languages and 50 dialects [8]. ElevenLabs introduced Realtime speech-to-speech models covering over 90 languages [8]. Block previewed Dessa, a local full-duplex model trained exclusively on human speech [9]. Despite these advancements, limitations persist; PolyAI CTO Shawn Wen indicated that voice AI still requires faster reasoning to reach its "ChatGPT moment" [10]. Research from the week found that full-duplex systems may degrade in long conversations, and real-time models can sometimes miss response timing [8]. Safety and robustness are also becoming central considerations, with studies reporting speaker-triggered speech-recognition backdoors achieving 90.2% success on Whisper-medium and 97.0% on Parakeet [8]. Deepfake detectors showed substantial degradation when voice transmissions occurred over real phone lines [8].
Sources (10)
  1. 1Top Text to Speech AI this week Odds & Predictions - Public.compublic.com
  2. 2Top Text to Speech AI this week Odds & Predictions 2026kalshi.com
  3. 3Best Text-to-Speech APIs in 2026, Compared | DevToolLab Blogdevtoollab.com
  4. 4ElevenLabs vs OpenAI vs Gemini TTS: AI Voice Generator 2026tech-insider.org
  5. 5Best Text to Speech API 2026: 6 Compared, With Pricesforasoft.com
  6. 6Voice AI compared: speech-to-text, text-to-speech and voice agents from ElevenLabs, OpenAI, Google and 20 others (October 2026) - Dbggrdbggr.dev
  7. 7Amazon Polly delivers a major upgrade to its Generative engine | Artificial Intelligenceaws.amazon.com
  8. 8Voice AI News - Week 43, Oct 19-25, 2026 | VoiceAISpacevoiceaispace.com
  9. 9Block's new voice model sounds eerily human | TechCrunchtechcrunch.com
  10. 10These execs think voice AI hasn't reached its ChatGPT moment yet | TechCrunchtechcrunch.com

2. Price Chart

Historical Price (Probability)

Outcome probability
Date

3. Significant Price Movements

Notable price changes detected in the chart, along with research into what caused each movement.

📉 October 26, 2026: 87.0pp drop

Price decreased from 88.0% to 1.0%

Outcome: ElevenLabs

What happened: Despite an 87.0 percentage point drop for "ElevenLabs" in the "Top Text to Speech AI this week" market on October 26, 2026, the provided research does not indicate a clear primary driver for this decline. Social media activity and traditional news around this date primarily highlighted positive developments, such as the October 20–21 launch of Eleven Realtime speech-to-speech models and news attention on October 26 for the Instant Soundtracks AI music tool [1][2][3][4]. No negative social media posts from key figures or viral narratives coinciding with or leading the price drop are substantiated by the available social snapshot [1]. Therefore, social media appears irrelevant or, at least, not a primary driver for this market movement based on the evidence.

📈 October 25, 2026: 84.0pp spike

Price increased from 4.0% to 88.0%

Outcome: ElevenLabs

What happened: The primary driver of the ElevenLabs prediction market price movement was a significant surge in social media activity and mentions, which LunarCrush reported as an 84.16% week-over-week increase across platforms [5]. This attention spike coincided with ElevenLabs' announcement of Chimera, a workflow combining Eleven v4 audio with Seedance 2.5 video for animation and lip sync, shared by @elevenlabs around October 25 [6]. Concurrent discussions also highlighted Eleven v4's practical advantages and stronger expressive delivery [7][8]. Social media was a primary driver, reflecting accumulated buzz and amplifying ElevenLabs' leading position in the text-to-speech AI market [5][9].
Sources (9)
  1. 1ElevenLabs on X / Twitterplushcap.com
  2. 2ElevenLabs Launches Eleven Realtime Speech-to-Speech Models | Petagenius Newsnews.petagenius.com
  3. 3ElevenLabs launches Instant Soundtracks to generate original music from uploaded videosdigitaltoday.co.kr
  4. 4ElevenLabs debuts AI tool that auto-scores videos with music and effects - CHOSUNBIZbiz.chosun.com
  5. 5Topic elevenlabs-ai | LunarCrush for AIlunarcrush.ai
  6. 6Introducing Chimera. Chimera pairs Eleven v4 audio with Seedance 2.5… — reel by @elevenlabs | PicManypicmany.com
  7. 7ElevenLabs v4 vs v3 (2026): Same Scripts, Listen to Bothaitoolwise.com
  8. 8Eleven v4 audio tags, IPA and dialogue explainedinfosdivers.com
  9. 9ElevenLabs: is it rising? (Steady, trend 54) · SoonViralsoonviral.com

4. Market Data

Contract Snapshot

This market concerns which Text-to-Speech AI, among SpeechifyAI, StepFun, and VUI Labs, will be deemed "Top Text to Speech AI this week." The page does not specify the exact criteria for determining a YES or NO resolution for any of the listed companies. The market begins and resolves on October 30, 2026, at 12:00 am EDT, with the maximum payout also on this date.

Available Contracts

Market options and current pricing

Outcome bucket Yes (price) No (price) Last trade probability
Cartesia $0.79 $1.00 11%
Async $0.79 $1.00 10%
Google $0.79 $1.00 10%
Inworld $0.79 $1.00 10%
MiniMax $0.79 $1.00 10%
Mistral $0.79 $1.00 10%
Smallest.ai $0.84 $1.00 10%
Alibaba $0.74 $1.00 9%
SpaceXAI $0.78 $1.00 9%
Nvidia $0.69 $1.00 8%
SpeechifyAI $0.01 $1.00 5%
StepFun $0.01 $1.00 5%
VUI Labs $0.01 $1.00 5%
Amazon $0.79 $1.00 3%
ElevenLabs $0.82 $1.00 1%
Fish Audio $0.01 $1.00 1%
OpenAI $0.01 $1.00 1%

Market Discussion

The week of October 19–25, 2026 saw a notable shift toward real-time, expressive, full-duplex voice systems, with Mistral launching Voxtral 3 and ElevenLabs launching Eleven Realtime for conversational agents, and Block previewing Dessa as a research direction [1][2][3]. For pure expressive scripted text-to-speech, Fish Audio Drama 3 is a leading candidate, available October 13, 2026, while Gradium's Phonon stands out for open-source, on-device deployment [4][5].

Sources (5)
  1. 1Voice AI News - Week 43, Oct 19-25, 2026 | VoiceAISpacevoiceaispace.com
  2. 2Block's new voice model sounds eerily human | TechCrunchtechcrunch.com
  3. 3Block, ElevenLabs, IBM and Ant Ship New Voice Modelsvoice-ai-newsletter.krisp.ai
  4. 4Fish Audio Drama 3: The Most Emotive Text-to-Speech Model We've Built - Fish Audio Blogfish.audio
  5. 5Gradium releases Phonon, open-source on-device Text-to-Speech | Gradiumgradium.ai

5. Trust Index

Octagon Trust Index Kalshi 69 Caution

Primary risk· Trade quality

How it adds up
Integrity80% of score78Good

Market integrity is low (65), but Integrity averages all three scores, so the other two pull it up. Only a critically low score would cap the total.

Trade quality20% of score34High Risk
Trust score69Caution

Weighted blend with hard caps — a critically weak safety pillar, or a severe trading anomaly, caps the total regardless of the rest. Full methodology · About the Trust Index

Trust profile
Integrity4 screens run · 6 don't apply
Market integrity
65Caution

Outcome can be influenced

Thin-volume movesTrade signals · high confidence

“KXSPEECHAI 26OCT30 CART” made a sharp jump with almost no trading behind it.

Screened clean
Large tradesNo single trade dominates the tape. Clear
Size patternsTape too thin Clear
2 screens not run
Outcome control · doesn't apply to this market type
Thin-leg conviction · doesn't apply to this market type
Info fairness
80Good

Structural prior — no screens apply

No suspicious activity on any screen

Screened clean
Information exposureNo concentrated group can know the outcome materially early. Clear
4 screens not run
Move vs. catalyst · doesn't apply to this market type
Longshot repricing · doesn't apply to this market type
Pre-broadcast certainty · doesn't apply to this market type
Cross-venue timing · doesn't apply to this market type
Resolution quality
92Strong

Resolution mechanism is dependable

No suspicious activity on any screen

Resolution rules read: Clear

Liquidity
33Very thin

Hard to fill near the screen price

2 contracts · worst first · 15 below the volume floor — not scored

ElevenLabs30
Very thin · low confidence

Light recent trading — size may not fill near the screen price.

Amazon37
Very thin · low confidence

Light recent trading — size may not fill near the screen price.

Move quality
54Mixed

Some moves lacked volume

2 contracts · worst first · 15 below the volume floor — not scored

Amazon36
Mixed · high confidence

Volume is light/neutral for the move over 6h

ElevenLabs71
Confirmed · medium confidence

Volume is light/neutral for the move over 24h

Rule clarity
100Clear

Counts toward Resolution quality, not here

2 contracts · worst first · 15 below the volume floor — not scored

Amazon100
Clear · high confidence

All checks pass

objective outcome · Clearofficial source · Cleartime boundary · Clear
ElevenLabs100
Clear · high confidence

All checks pass

objective outcome · Clearofficial source · Cleartime boundary · Clear
Trade quality
34High Risk

Thin market — entering or exiting at fair value will be difficult.

Outcome quality 34 (notional-weighted across 2 scored legs)

15 below the volume floor — not scored

AmazonPrimary37Market scoreLiquidity37Move quality36Rule clarity100
Why 37
Liquidity
37 · Very thin · low confidence

Light recent trading — size may not fill near the screen price.

Move quality
36 · Mixed · high confidence

Volume is light/neutral for the move over 6h

Rule clarity
100 · Clear · high confidence

All checks pass

objective outcome · Clearofficial source · Cleartime boundary · Clear
ElevenLabs30Market scoreLiquidity30Move quality71Rule clarity100
Why 30
Liquidity
30 · Very thin · low confidence

Light recent trading — size may not fill near the screen price.

Move quality
71 · Confirmed · medium confidence

Volume is light/neutral for the move over 24h

Rule clarity
100 · Clear · high confidence

All checks pass

objective outcome · Clearofficial source · Cleartime boundary · Clear

underwriting_v1.4 · computed Oct 27, 2026

6. What specific metrics does the Artificial Analysis Speech Arena leaderboard use to rank models, and how might its methodology favor certain providers like Cartesia or Smallest.ai?

Cartesia Sonic 3.5 (Controlled Voice Arena)1,122 Elo [1][2]
Cartesia Sonic 3.6 (Provider Voice Board)#4 with 1,292 Elo [3]
Smallest.ai (Provider Voice Arena)28th of 103 [4][5]
Artificial Analysis ranks models using a human preference-based Elo system. This leaderboard employs a quality/preference Elo system, where listeners create rankings through blind binary pairwise choices between audio clips [4][6][7]. This methodology leverages a Bradley–Terry maximum-likelihood model, anchoring one model at 1000 Elo and prioritizing subjective human preference for attributes such as prosody and naturalness, rather than objective metrics like speed, price, or throughput [4][2]. The Elo estimation system is recomputed daily from the complete vote history, ensuring past votes remain influential, while also favoring recent entrants in matchup sampling and down-weighting frequently repeated pairings [4].
Leaderboard design and comprehensive prompts influence provider performance and evaluation. The Arena's specific design choices, including distinct boards like Provider Voice for curated native voices and Controlled Voice for evaluating cloning quality with a shared reference, can affect which providers are favored [4][6][2]. Prompt coverage is extensive, encompassing challenging alphanumeric and numeric sequences across various domains, which are standardized and loudness-normalized before evaluation [4].
Cartesia shows strong performance; Smallest.ai's ranking suggests less favorability. As of the cited materials from 2026, Cartesia demonstrated strong performance under Artificial Analysis's preference methodology. Cartesia Sonic 3.5 led the Controlled Voice Arena with 1,122 Elo, and its Sonic family led multiple new language boards [1][2]. An excerpt from the current Provider Voice board places Cartesia Sonic 3.6 at #4 with 1,292 Elo [3]. Smallest.ai, while potentially benefiting if its strengths align with perceived naturalness, is currently ranked 28th of 103 on the Provider Voice Arena preference Elo, according to its Artificial Analysis page [4][5]. Therefore, for a prediction market resolving October 30, 2026, Cartesia's prospects improve if "Top Text to Speech AI" refers to Controlled Voice or a specific language board, but Smallest.ai is not indicated as the likely overall winner based on the provided snapshot [4][1][2][3].
Sources (7)
  1. 1Multilingual TTS Leaderboards for 9 Languages Releasedlinkedin.com
  2. 2Cartesia Sonic 3.5 leads in Controlled Voice Arena...linkedin.com
  3. 3Text to Speech Leaderboard - Top AI Speech Models | Artificial Analysisartificialanalysis.ai
  4. 4Text to Speech Benchmarking Methodologyartificialanalysis.ai
  5. 5Smallest.ai - Quality Elo, Speed & Price Analysisartificialanalysis.ai
  6. 6Speech Arena - Top AI Speech Models | Artificial Analysisartificialanalysis.ai
  7. 7Speech Arena | Artificial Analysisartificialanalysis.ai

7. What recent model updates from Mistral (Voxtral 3) or Amazon (Polly Generative) could significantly alter their leaderboard rankings before the October 30 resolution?

Voxtral 3 Current Rank#12 (Elo 1221±14) [1]
Polly Generative Current Rank#59 (Elo 1063±11) [1]
Voxtral 3 Release DateOctober 20–21, 2026 [2][3][4]
Mistral's Voxtral 3 shows significant potential for leaderboard advancement. Mistral's Voxtral 3, launched between October 20–21, 2026, is currently positioned at #12 on the leaderboard with an Elo score of 1221±14 [1][2][3][4]. This places it materially closer to the top, and its release, alongside a LiveKit-based voice-agent SDK, is assessed as having a realistic near-term ranking-upside story, suggesting a plausible short-term leaderboard catalyst [1][2][3][4][5][6]. The model reports capabilities including approximately 90 milliseconds time to first audio, support for 34 emotion tags, 2 pace tags, and claims of broader multilingual and dialect support [2][3][4].
Amazon's Polly Generative upgrade did not immediately impact its leaderboard standing. Conversely, Amazon's Polly Generative received an upgrade between October 20–22, 2026, yet its leaderboard position remains at #59 with an Elo score of 1063±11 [1][5][6][7][8]. This upgrade, which included a new model, more natural speech, and an expansion of voices and languages, is characterized as a production and catalog expansion rather than an indication of a new evaluated leaderboard score [5][6][7][8]. The leaderboard still displays Polly Generative at #59 and attributes its release to May 2024, making a jump to the #1 position by the October 30 resolution date substantially less plausible without a dramatic Arena re-evaluation [1][5][6][9].
Sources (9)
  1. 1Text to Speech Leaderboard - Top AI Speech Modelsartificialanalysis.ai
  2. 2Mistral launches Voxtral 3 speech models and a voice...daily.dev
  3. 3Mistral AI Releases Voxtral 3 Speech Models and Agent SDKemergent.sh
  4. 4Voxtral TTS 3 Context, Specs & Sources (October 2026)benchlm.ai
  5. 5Polly Generative engine upgraded with more natural...aws.amazon.com
  6. 6Amazon Polly delivers a major upgrade to its Generative...aws.amazon.com
  7. 7AWS Weekly Roundup: Amazon Polly Generative engine...aws.amazon.com
  8. 8Document History for Amazon Pollydocs.aws.amazon.com
  9. 9Top Text to Speech AI this week Odds & Predictions 2026kalshi.com

8. What evidence explains the discrepancy between ElevenLabs' high qualitative Elo ratings and its low implied probability in the prediction market?

ElevenLabs Elo Rating1333–1338 (Oct. 26–27, 2026) [1][2]
Artificial Analysis Rank#1 (Oct. 26–27, 2026) [1][2]
Market Implied Probability Range1% to 88% across venues (Oct. 30 market) [3][4]
The significant discrepancy between ElevenLabs' high qualitative Elo ratings and its low implied probability in prediction markets is primarily attributed to inconsistent and sparse market data. Evidence from October 27 does not conclusively establish a low probability, as different venues display materially disagreeing figures, suggesting stale, crossed, or platform-specific market feeds [3][4].
ElevenLabs models hold top Elo ratings, but their lead is narrow. As of October 26–27, 2026, Eleven v4 Turbo and Eleven v4 hold top or near-top positions in the Artificial Analysis Provider Voice Speech Arena, with Elo ratings ranging from approximately 1333 to 1338, based on blind pairwise listener evaluations [1][2]. However, this qualitative leadership is narrow and statistically uncertain, with close competitors such as Qwen-Audio, Sonic, and Gemini trailing by only tens of Elo points [1][5][2]. The market scheduled for October 30 will resolve based on a future snapshot of the Artificial Analysis Speech Arena ranking, making a specific, point-in-time leaderboard position the key tradable event [6][3].
Discrepant market probabilities suggest event risk, not quality rejection. Public's snapshot currently estimates ElevenLabs at 1% despite an 81¢ 'Yes' price, which contrasts sharply with a Frenzy aggregation showing 50%, MiniMax at 48%, and the ElevenLabs-specific Frenzy page reporting 88% [3][4]. If a truly low implied probability were clearly established, the strongest explanation would be event-risk rather than a rejection of ElevenLabs' current quality. This risk factors in future rank volatility, tight Elo confidence intervals, potential ties, changes in leaderboard methodology, and intense competition from several nearly leading providers [6][3][4]. Furthermore, historical weekly markets utilizing the same Artificial Analysis source have demonstrated significant reversals and non-ElevenLabs winners, including Cartesia, which resolved at 99% in September [7][8][9].
Sources (9)
  1. 1Text to Speech Leaderboard - Top AI Speech Modelsartificialanalysis.ai
  2. 2TTS Leaderboard — Best Text-to-Speech Models (Elo)pandanpc.com
  3. 3Top Text to Speech AI this week Odds & Predictions - Public.compublic.com
  4. 4Top Text to Speech AI this week?… | Frenzy Capitalfrenzycap.com
  5. 5ElevenLabs' Eleven v4 Turbo Tops TTS Leaderboard at...alphasignal.ai
  6. 6Top Text to Speech AI this week Odds & Predictions 2026kalshi.com
  7. 7Top Text to Speech AI this week Odds | PredictPediapredictpedia.com
  8. 8Top Text to Speech AI this week - Resultado: Cartesia | CoinRithmcoinrithm.com
  9. 9Top Text to Speech AI this week - Result: Cartesia | CoinRithmcoinrithm.com

9. How do market front-runners Cartesia and ElevenLabs compare on low-latency streaming performance for real-time applications?

Cartesia Sonic 3.5 Median Latency (Vapi)128 ms median [1]
ElevenLabs Flash v2.5 Median Latency (Vapi)197 ms median live-streaming [2]
ElevenLabs Flash v2.5 Advertised Inference Latencyapproximately 75 ms (excluding overhead) [3]
Cartesia and ElevenLabs both deliver strong low-latency streaming performance for real-time applications. Independent measurements show varied results depending on the specific model version and testing methodology [4][2][5]. Cartesia is often considered a low-latency front-runner for interactive agents using Sonic 3.5 or equivalent fast endpoints, with ElevenLabs serving as a credible challenger [4]. However, the ranking is materially affected by factors such as model version, endpoint, geography, connection reuse, and measurement definitions, indicating that neither company is unambiguously dominant across all tests [4][5][1][2].
Cartesia's Sonic 3.5 model shows diverse latency measurements across different reports. Cartesia/Coval reported Sonic 3.5 at 351 ms P90 and 320 ms median, while its product documentation claims sub-90 ms model latency [4][6]. AWS described Sonic 3 as sub-100 ms, though these figures are not directly comparable due to differing exclusions of network and application overhead [7][4]. An independent Vapi benchmark reported Cartesia Sonic 3.5 at 128 ms median [1]. Conversely, a September 2026 Coval read, cited by Gradium, reported Cartesia Sonic 3.6 at 440 ms [5]. Cartesia supports both WebSocket and REST protocols, provides SSM-based streaming, and offers extensive real-time controls, including playback before full synthesis, emotion, volume, and speed adjustments [4][6][7]. It supports 42–44 languages, depending on the model and source [4][6][7].
ElevenLabs' Flash v2.5 is optimized for real-time applications, with varied latency reports. It advertises approximately 75 ms inference latency, which excludes application and network latency [3]. ElevenLabs documents regional WebSocket time-to-first-audio (TTFB) figures, ranging roughly from 100–150 ms in North America, Europe, and Southeast Asia, to 150–200 ms in South and Northeast Asia [3]. An independent Vapi benchmark indicated 197 ms median live-streaming latency for Flash v2.5, which includes network RTT [2]. However, a September 2026 Coval read, cited by Gradium, reported Flash v2.5 at 185 ms median [5]. ElevenLabs supports progressive HTTP streaming in addition to bidirectional WebSockets, but it warns that factors such as chunk scheduling, voice choice, output format, geography, and connection reuse can significantly alter time-to-first-audio [3][8]. Flash v2.5 supports 32 languages, trading some quality and emotional depth for speed, with higher-expressiveness models being slower [8].
Sources (8)
  1. 1Cartesia Sonic 3.5: Humanness & Latency | Vapihumannessindex.vapi.ai
  2. 2ElevenLabs Flash v2.5: Humanness & Latency | Vapihumannessindex.vapi.ai
  3. 3Latency optimization | ElevenLabs Documentationelevenlabs.io
  4. 4ElevenLabs alternative for voice agentscartesia.ai
  5. 5Best Low-Latency TTS APIs 2026gradium.ai
  6. 6Sonic AI voice: real-time TTS API with emotioncartesia.ai
  7. 7Cartesia Sonic 3 text-to-speech model is now available on...aws.amazon.com
  8. 8Models | ElevenLabs Documentationelevenlabs.io

10. As the frontier shifts to conversational AI, do Block's Dessa or Inworld's models show superior full-duplex performance that could act as a tie-breaker on advanced leaderboards?

Market Resolution DateOct. 30, 2026 [1]
Inworld Realtime TTS-2 Ranking#6 with 1,271 Speech Arena Elo [2]
Block Dessa Public AvailabilityNo public API, weights, or benchmark score available [3][4][5][6][7]
Neither Dessa nor Inworld's full-duplex capabilities can break current TTS leaderboard ties. Their full-duplex performance does not serve as a credible tie-breaker for the "Top Text to Speech AI this week" prediction market because the market's resolution, set for October 30, 2026, is specifically linked to the designated Artificial Analysis Speech Arena leaderboard, not to any unpublished full-duplex benchmarks [1]. Full-duplex leaderboards typically evaluate operational conversational behaviors, such as pause handling and turn-taking, extending beyond merely producing compelling demo audio [8][9].
Block's Dessa, though promising, lacks public metrics for leaderboard influence. Dessa is currently described as a research preview of full-duplex speech-to-speech technology. However, as of October 27, 2026, public information indicates no available public API, weights, independent evaluation, latency figures, or quantitative benchmark scores for Dessa [3][4][5][6][7]. For Dessa to influence the market's designated leaderboard, it would need to undergo public evaluation and be listed before the market's cutoff date, which currently appears unlikely [3][7][8][2].
Inworld Realtime shows some strength but does not dominate current TTS rankings. Inworld Realtime is an evaluated and deployable full-duplex stack, with the company asserting its voice technology is sub-second and ranked #1 on Artificial Analysis's realtime voice benchmark [8][10]. This claim is vendor-reported and does not establish overall full-duplex dominance [8]. Current TTS data from an October 26 aggregator lists Inworld Realtime TTS-2 at #6 with 1,271 Speech Arena Elo [2]. While Artificial Analysis reported Inworld Realtime TTS-2 led Mandarin with 1,185 Elo, Cartesia's Sonic family led 8 of 9 other newly reported languages [11]. Inworld can impact this market through its existing TTS entries, yet full-duplex quality is not the stated resolution criterion for this market, and it currently does not displace leading established TTS candidates like ElevenLabs, MiniMax, and Cartesia on the Artificial Analysis Speech Arena leaderboard [3][7][8][2][1][12][13].
Sources (13)
  1. 1Top Text to Speech AI this week Odds & Predictions 2026kalshi.com
  2. 2TTS Leaderboard — Best Text-to-Speech Models (Elo)pandanpc.com
  3. 3Dessa Context, Specs & Sources (October 2026)benchlm.ai
  4. 4Block Previews Dessa Voice AI Trained Without Texttechrepublic.com
  5. 5Dessa Full-Duplex Speech Model: 0 Text Tokens [2026]shattered.io
  6. 6Block Dessa: The Speech Model Trained Without Textcellcog.ai
  7. 7Dessa: Block's Unreleased Full-Duplex Voice AI, Explainedmindstudio.ai
  8. 8Speech to Speech Models and Providers Analysisartificialanalysis.ai
  9. 9Benchmarks - Fullduplexfullduplex.ai
  10. 10Build voice agents that sound like real peopleinworld.ai
  11. 11Multilingual TTS Leaderboards for 9 Languages Releasedlinkedin.com
  12. 12Top Text to Speech AI this week — Live Odds | Frenzy Capitalfrenzycap.com
  13. 13Top Text to Speech AI this week Odds & Predictions - Public.compublic.com

11. What Could Change the Odds

Key Catalysts

Prediction markets for late October 2026 identify several key AI and technology catalysts. These include company rankings for 'best AI model', GPU rental price trends, and the status of ongoing regulatory inquiries by the EU AI Office [1][2].
Markets are currently bearish on significant new model releases from major labs like OpenAI and Fable before the end of October 2026, with most probability focused on maintaining existing market positions or minor updates [1][2]. As of October 27, 2026, Eleven v4 Turbo and Eleven v4 consistently rank as the leading text-to-speech models on major benchmarks, followed by Google's Gemini 3.8 Flash TTS [3][4].
Major prediction markets, including Polymarket, show heavy volume around 'best AI model' and 'best AI lab' titles as of October 31, 2026 [1][5]. Google and Anthropic are frequently cited as the frontrunners in these categories [1][5].

Key Dates & Catalysts

  • Expiration: October 30, 2026
  • Closes: October 30, 2026
Sources (5)
  1. 1October 2026 Prediction Market Calendar: 1140 Polymarket Markets by Datepolymarkettrader.com
  2. 2AI and Tech Prediction Markets: What Traders Are Pricing This Week (October 2026) – PredictWirepredictwire.io
  3. 3Text to Speech Leaderboard - Top AI Speech Modelsartificialanalysis.ai
  4. 4Best Text-to-Speech Models — AI Voice Generator Models Ranked (October 2026) · modelgrepmodelgrep.com
  5. 5#3 AI Lab end of October? (Style Control On)polymarket.com

13. Related News

-87.0pp
Last updated: October 27, 2026, 12:19 UTC

Mistral TTS Launch Upsets AI Voice Market, Shattering ElevenLabs' Lead

The recent release of Mistral's "Voxtral 3" text-to-speech (TTS) model has triggered a significant repricing in the market for the top TTS AI, with odds for the long-time front-runner, ElevenLabs, col...

14. Historical Resolutions

Historical Resolutions: 20 markets in this series

Outcomes: 1 resolved YES, 19 resolved NO

Recent resolutions:

  • KXSPEECHAI-26OCT23-VUI: NO (Oct 23, 2026)
  • KXSPEECHAI-26OCT23-STEP: NO (Oct 23, 2026)
  • KXSPEECHAI-26OCT23-SPEE: NO (Oct 23, 2026)
  • KXSPEECHAI-26OCT23-SPAC: NO (Oct 23, 2026)
  • KXSPEECHAI-26OCT23-SMAL: NO (Oct 23, 2026)