Charted: AI Models Tied for First in Voice and Video Cost 7× to 12× Apart

Pick the best-rated AI video generator on a blind-vote leaderboard and you will find it is not alone. Artificial Analysis' text-to-video arena ranks 29 models with native audio on social-media and creator content. The leader, Dreamina Seedance 2.5, scores 1188 Elo with a 95% interval of 1158 to 1218. Six other models have intervals that overlap it, so the votes cannot reliably separate them from the leader. Six of the seven carry a tracked price, and those prices run from $4.80 to $34.12 per minute of finished video: a 7.1× spread for results that voters could not tell apart.

Voice shows the same pattern, more sharply. On the speech arena's entertainment board, seven of 89 models tie for first, and all seven have a tracked price, from $0.0066 per minute (Simba 3.2) to $0.08 per minute (Eleven v4). That is a 12× spread. Across the 75 voice models that have both a price and an entertainment score, the rank correlation between price and quality is 0.12, close to no relationship. Video prices carry more signal (0.45 across 27 models), but the top of the board is crowded there too.

What the numbers are

Everything here comes from a database built on October 11, 2026. It converts every price to US dollars per minute and sets it beside published quality scores: 265 models, 639 price rows (366 read from providers' own pages, 232 of those subscription or credit plans, and 273 tracked by Artificial Analysis or LMArena) and 4,661 scores from 13 benchmarks. Quality means blind human preference: listeners or viewers pick the better of two outputs, and the votes become Elo-style scores with a 95% interval. Scores from different boards are not comparable, so each figure compares models inside a single board. “Statistically tied” below means a model's interval overlaps the leader's.

Video: seven models share the top, at prices 7× apart

Figure 1 (above) plots each priced model's social-media score against its tracked price per minute of video with native audio, on a log scale. The dashed line joins the best-value frontier: the models that no other model beats on both score and price. It has four members: Agnes-Video-2.5 ($1.50 per minute, 1089 Elo), MiniMax H3 at 768p ($4.80, 1155), Wan 3.0 ($12.00, 1169) and Seedance 2.5 ($34.12, 1188).

The top of that line is the expensive part. Moving from MiniMax H3 to Seedance 2.5 costs 7.1× as much for 33 more Elo points. At that gap the Elo formula implies Seedance 2.5 would win about 55% of head-to-head votes against MiniMax H3, so a real edge, but a small one, and the two intervals overlap. Wan 3.0 sits in between: 19 points below the leader at 35% of its tracked price.

Two of these tracked prices disagree with the provider's own page. Seedance 2.5 is tracked at $34.12 per minute while BytePlus lists $18.18 at 720p; at the lower figure the premium over MiniMax H3 shrinks from 7.1× to 3.8×. MiniMax H3 Max is tracked at $4.80 and MiniMax's page lists $3.00 at 480p. Confirm the resolution you need before budgeting.

Voice: price barely predicts what listeners prefer

Show the data behind this figure

Model	Creator	Elo	95% interval	Tied with #1	Tracked $/min
Eleven v4	ElevenLabs	1287	1241 to 1333	yes	$0.08
Eleven v4 Turbo	ElevenLabs	1286	1235 to 1337	yes	$0.04
Qwen-Audio-3.1-TTS-Plus	Alibaba	1250	1203 to 1297	yes	$0.0193
Gemini 3.8 Flash TTS	Google	1249	1208 to 1290	yes	$0.0165
Qwen-Audio-3.0-TTS-Plus	Alibaba	1230	1190 to 1270	yes	$0.0193
Simba 3.2	SpeechifyAI	1214	1179 to 1249	yes	$0.0066
Sonic 3.6	Cartesia	1208	1168 to 1248	yes	$0.049
v3 Conversational	ElevenLabs	1192	1154 to 1230		$0.05
Realtime TTS-2	Inworld	1186	1144 to 1228		$0.0208
Gemini 3.8 Flash-Lite TTS	Google	1186	1148 to 1224		$0.011
Gemini 3.1 Flash TTS	Google	1184	1152 to 1214		$0.0183
Luna TTS	VUI Labs	1163	1130 to 1196		$0.015
Realtime TTS-2 Flash	Inworld	1159	1121 to 1197		$0.0104
StepAudio 2.5 TTS (Aug 2026)	StepFun	1155	1116 to 1194		$0.085
Sonic 3.5	Cartesia	1135	1103 to 1167		$0.049
Breeze TTS 2	BreezeBlue	1128	1090 to 1166		$0.034
Speech 2.8 HD	MiniMax	1127	1098 to 1156		$0.10
Eleven v3	ElevenLabs	1126	1097 to 1155		$0.10
Soniox TTS Real-Time v2	Soniox	1121	1092 to 1150		$0.0142
Speech 2.8 Turbo	MiniMax	1108	1080 to 1136		$0.06
Step TTS 2 (Mar 2026)	StepFun	1096	1063 to 1129		$0.04
Lightning V3.1 Pro (Jul 2026)	Smallest.ai	1086	1052 to 1120		$0.0195
SpaceXAI TTS	SpaceXAI	1081	1042 to 1120		$0.015
Fish Audio S2.1 Pro	Fish Audio	1081	1047 to 1115		$0.015
Gradium TTS (Aug 2026)	Gradium	1079	1044 to 1114		$0.0472
Speech 2.6 Turbo	MiniMax	1077	1048 to 1106		$0.06
Falcon 2	Murf AI	1075	1039 to 1111		$0.01
Fish Audio S2 Pro	Fish Audio	1075	1042 to 1108		$0.015
Azure HD 2.5	Microsoft	1074	1043 to 1105		$0.022
Gradium TTS (Sep 2026)	Gradium	1073	1034 to 1112		$0.0478
Async Flash v1.5	async	1067	1034 to 1100		$0.0101
Simba 3.0	SpeechifyAI	1067	1036 to 1098		$0.0066
Speech-02-HD	MiniMax	1067	1035 to 1099		$0.10
Bland Speech v3	Bland AI	1058	1024 to 1092		$0.04
Speech 2.6 HD	MiniMax	1056	1027 to 1085		$0.10
TTS-1 HD	OpenAI	1053	1023 to 1083		$0.03
Turbo v2.5	ElevenLabs	1051	1022 to 1080		$0.05
Speech-02-Turbo	MiniMax	1048	1015 to 1081		$0.06
Gemini 2.5 Flash Lite TTS	Google	1046	1014 to 1078		$0.0092
TTS-1	OpenAI	1044	1014 to 1074		$0.015
T2A-01-HD	MiniMax	1044	1012 to 1076		$0.05
GPT-Realtime-2	OpenAI	1044	1009 to 1077		$0.19
Async Pro v1.0	async	1042	1008 to 1076		$0.0202
OpenAudio S1	Fish Audio	1039	1007 to 1071		$0.015
Voxtral TTS	Mistral	1037	1003 to 1071		$0.016
Flash v2.5	ElevenLabs	1036	1006 to 1066		$0.05
Sonic 3	Cartesia	1029	998 to 1060		$0.049
Flux TTS	Deepgram	1028	991 to 1065		$0.045
Multilingual v2	ElevenLabs	1026	996 to 1056		$0.10
Studio	Google	1020	988 to 1054		$0.16
Journey	Google	1019	985 to 1053		$0.16
Gemini 2.5 Flash TTS (Dec 2025)	Google	1015	983 to 1047		$0.0092
Octave TTS	Hume AI	1013	980 to 1046		$0.0875
Chatterbox HD	Resemble AI	1011	975 to 1047		$0.04
Higgs Audio V3 TTS	Boson AI	1008	973 to 1043		$0.015
Coda	Rime	1002	968 to 1036		$0.05
Zonos-v0.1	Zyphra	1000	1000 to 1000		$0.02
Octave 2	Hume AI	999	967 to 1031		$0.0875
Chirp 3: HD	Google	997	965 to 1029		$0.03
Polly Generative	Amazon	994	959 to 1029		$0.03
Async Flash v1.0	async	993	963 to 1023		$0.0101
Kokoro 82M v1.0	Kokoro	992	961 to 1023		$0.0006
Polly Long-Form	Amazon	985	950 to 1020		$0.10
Lightning v3.1	Smallest.ai	984	950 to 1018		$0.025
T2A-01-Turbo	MiniMax	963	929 to 997		$0.03
Chatterbox	Resemble AI	961	928 to 994		$0.025
Qwen3 TTS	Alibaba	924	890 to 958		$0.0138
Murf Speech Gen 2	Murf AI	909	873 to 945		$0.10
Qwen3 TTS Flash	Alibaba	877	838 to 916		$0.011
WaveNet	Google	872	836 to 908		$0.004
Standard	Google	870	833 to 907		$0.004
OpenVoice v2	OpenVoice	838	793 to 883		$0.0083
Neural2	Google	828	790 to 868		$0.016
Polly Standard	Amazon	819	773 to 865		$0.004
Polly Neural	Amazon	816	775 to 857		$0.016

Figure 2 repeats the exercise for speech. The seven models tied with the leader are Simba 3.2 ($0.0066 per minute), Gemini 3.8 Flash TTS ($0.0165), Qwen-Audio-3.1-TTS-Plus and Qwen-Audio-3.0-TTS-Plus ($0.0193 each), Eleven v4 Turbo ($0.04), Sonic 3.6 ($0.049) and Eleven v4 ($0.08). The cleanest example sits inside one product family. Eleven v4 and Eleven v4 Turbo score 1287 and 1286 on the entertainment board, 1308 and 1310 on knowledge sharing, and 1323 and 1327 overall: the same quality within noise. The standard model is tracked at $0.08 per minute and the Turbo variant at $0.04, and ElevenLabs' pricing page carries the same two figures.

Listener preference is not the only thing worth testing. Artificial Analysis also runs a pronunciation test in which reviewers judge whether hard items such as numbers, codes and brand names are said correctly. Among the tied models that have a pass rate, scores range from 74.5% to 91.7%, and price does not order them: Sonic 3.6, the second most expensive, has the lowest.

Simba 3.2 ties the leader on listener preference at $0.0066 per minute but passes 77.6% of pronunciation checks, against 90.1% for Eleven v4 Turbo at about six times the price. The two measures agree only moderately (rank correlation 0.67 across 28 shared models), so a cheap voice that ties on taste can still miss more hard items in a script full of numbers. The best trade in the table is Gemini 3.8 Flash TTS: 89.5% at the tracked $0.0165, within a point of Eleven v4 Turbo at under half its price.

Voice prices also move. Google lists Gemini 3.8 Flash TTS at $9 per million audio tokens through December 31, 2026 and $18 from January 1, 2027. At 25 audio tokens per second that is about $0.0135 per minute now and $0.027 after the increase, against the $0.0165 tracked.

Music: the leader is also the cheapest, among the four with a price

Artificial Analysis runs two music arenas, one for instrumentals and one for tracks with vocals, covering 19 models. The arenas publish no prices, so every music price here comes from providers' own pages, and only four models could be priced. Most music tools sell credit plans rather than per-minute API rates, so a music price is the cheapest effective cost per minute across a provider's paid plans and API at full use of the allowance, free tiers excluded. A song is assumed to run three minutes.

Show the data behind this figure

Model	Creator	Instrumental	Vocals	$/min	$ per 3-minute song	Price basis
Suno v6	Suno	1140	1142	$0.0040	$0.012	Suno plan, full use of allowance
Suno v6-mini	Suno	1109	1116	no price found
Mureka V9	Mureka	1081	1052	no price found
Mureka V9.5	Mureka	1080	1103	no price found
Lyria 3 Pro	Google	1078	1046	$0.0267	$0.080	Google listed price
Lyria 3.5	Google	1076	1052	$0.0267	$0.080	Google listed price
Eleven Music	ElevenLabs	1021	1001	$0.1485	$0.446	ElevenLabs plan, full use of allowance
Stable Audio 3 Large	Stability.ai	1016		no price found
Eleven Music v2	ElevenLabs	1010	1001	no price found
MiniMax Music 2.5+	MiniMax	1005	988	no price found
MiniMax Music 2.6	MiniMax	1002	1001	no price found
Eleven Music v2.5	ElevenLabs	1002	1012	no price found
MiniMax Music 3.0	MiniMax	1000	1000	no price found
Stable Audio 3 Medium	Stability.ai	998		no price found
Melodia v3	Sonauto	988	978	no price found
Stable Audio 2.5	Stability.ai	986		no price found
Stable Audio 3 Small	Stability.ai	948		no price found
Lyria 2	Google	932		no price found
MusicGen	Meta	881		no price found

Of the four priced models, Suno v6 leads both boards (1140 instrumental, 1142 with vocals) at about $0.004 per minute, or $0.012 per three-minute song. Google's Lyria 3 Pro and Lyria 3.5 follow at 1078 and 1076 on the instrumental board and $0.08 per song through the API. Eleven Music scores 1021 at about $0.15 per minute ($0.446 per song). The other 15 models have no price here. That includes Mureka V9 and V9.5, which score between Suno and Lyria on the instrumental board. Their pages were blocked or script-only, or list no price, and MiniMax's paid music API is listed as discontinued. Commercial-use terms vary by plan, so check them before posting generated music on social platforms.

How far to trust the prices

Show all 19 models that differ by more than 8%

Model	Tracked $/min	Provider page $/min	Provider / tracked
Grok Imagine Video 1.5 Lite	$8.40	$1.20	0.14x
Dreamina Seedance 2.5	$34.12	$18.18	0.53x
Grok Imagine Video 1.5	$8.40	$4.80	0.57x
MiniMax H3 Max	$4.80	$3.00	0.62x
Gemini Omni Flash 1.1	$9.12	$6.00	0.66x
Lightning v3.1	$0.025	$0.0175	0.70x
Eleven v3	$0.10	$0.08	0.80x
Flash v2.5	$0.05	$0.04	0.80x
Multilingual v2	$0.10	$0.08	0.80x
Turbo v2.5	$0.05	$0.04	0.80x
v3 Conversational	$0.05	$0.04	0.80x
Veo 3.1 Fast	$9.00	$7.20	0.80x
Dreamina Seedance 2.0	$22.45	$18.18	0.81x
Gemini 3.8 Flash-Lite TTS	$0.011	$0.009	0.82x
Gemini 3.8 Flash TTS	$0.016	$0.0135	0.84x
Qwen3 TTS Flash	$0.011	$0.01	0.91x
Realtime TTS-2	$0.021	$0.025	1.19x
Realtime TTS-2 Flash	$0.01	$0.015	1.50x
Gemini 3.1 Flash TTS	$0.018	$0.03	1.67x

For 53 models the database holds both a tracked price and the provider's own. For 34 of them the two agree within 8%. For 19 they do not. Figure 4 shows the eight widest gaps as the provider's price divided by the tracked price. Provider pages are cheaper than the tracked price in 6 of those eight, and in 16 of all 19 disagreements, so budget from the provider's page for the exact resolution and tier you plan to use.

Do the benchmarks agree?

Rank correlation measures how similarly two boards order the models they share (1 means the same order). The video boards and Artificial Analysis' own slices agree closely, which is why a slice mostly re-ranks the overall order. The voice boards agree less, and the objective pronunciation test only partly follows preference.

One category has no independent score at all. The research behind this database found no independent, per-product benchmark for avatar and talking-head tools such as HeyGen, Synthesia or D-ID; what exists is vendor-run tests and academic benchmarks of research models. For avatar tools the database holds prices only, and quality has to be judged from sample clips.

Caveats and confidence

• Elo is relative. Scores are comparable only inside one board. Each figure here uses one board.
• “Tied” is a permissive test. Overlap with the leader's 95% interval says the votes cannot separate a model from the leader, not that the two are equal.
• Arenas measure typical prompts and raters, not your brand voice or niche. Re-test a shortlist on your own scripts.
• Tracked prices are list prices at default settings. Real cost depends on resolution, audio, plan and unused credits. Figure 4 shows where a provider's page disagrees.
• Music is thin. Four of 19 models are priced. Suno's price assumes a three-minute song and full use of a plan's credits, and unused credits raise it.
• Missing is not zero. Providers whose pages could not be read are absent from the price comparison. A $0 price, such as the one on MiMo-V2.5-TTS, is a promotion with no end date.
• Vintages differ. Artificial Analysis and Hugging Face data were captured on October 11, 2026; LMArena's vote cutoff is September 21, 2026. Prices change.

Bottom line for desks

In these arenas, a price is a weak guide to quality: nearly none in voice (0.12), modest in video (0.45). A better rule is to shortlist by the leader's interval, not the leader. Anything whose interval overlaps it is a candidate, and the cheapest candidate that passes your own test is the one to buy. On this snapshot that means tracked video prices from $4.80 per minute against $34.12, and voices from $0.0066 per minute against $0.08, subject to the pronunciation check for voice, the provider-page check for video, and a listen on your own material.

1. [AA Speech]Artificial Analysis — Text to Speech arena, provider-voice leaderboard (entertainment, knowledge-sharing and overall slices) and pronunciation robustness test. https://artificialanalysis.ai/text-to-speech/leaderboard/provider-voice
2. [AA Speech prices]Artificial Analysis — text-to-speech providers: tracked price per 1M characters, normalized here to USD per minute. https://artificialanalysis.ai/text-to-speech/providers
3. [AA Video]Artificial Analysis — text-to-video arena v2.0 (social-media and creator-content slice, with audio). https://artificialanalysis.ai/video/leaderboard/text-to-video
4. [AA Video prices]Artificial Analysis — video providers: tracked list price per minute. https://artificialanalysis.ai/video/providers
5. [AA Music]Artificial Analysis — music arena, instrumental and vocals leaderboards. https://artificialanalysis.ai/music/leaderboard/instrumental
6. [LMArena]LMArena (arena.ai) — text-to-video leaderboard, vote cutoff Sept 21, 2026, used for the benchmark-agreement table. https://arena.ai/leaderboard/video/text-to-video
7. [HF TTS Arena]TTS-AGI / Hugging Face — TTS Arena V2, used for the benchmark-agreement table. https://huggingface.co/spaces/TTS-AGI/TTS-Arena-V2
8. [BytePlus]BytePlus ModelArk billing — Seedance 2.5 and 2.0 provider price ($18.18 per minute at 720p). https://docs.byteplus.com/en/docs/byteplus_las/Large_model_billing
9. [Google AI pricing]Google — Gemini API pricing, Gemini TTS audio-token prices and the January 1, 2027 change. https://ai.google.dev/gemini-api/docs/pricing
10. [ElevenLabs]ElevenLabs — API pricing and plans (voice and Eleven Music). https://elevenlabs.io/pricing/api
11. [Suno]Suno — plans and credits, the basis for the Suno v6 effective price. https://suno.com/pricing
12. [MiniMax]MiniMax — pricing overview (H3 Max provider price; music API status). https://platform.minimax.io/docs/pricing/overview