1
Seedance 2.5
ByteDance · Directed, reference-heavy shots up to 30 seconds with sound
$$$$
The most consistent video model on the market: it is the only one in the top 7 on all six major public video arenas, and #1 on Artificial Analysis image-to-video. It has the deepest reference system of any model (up to 30 images, 10 videos and 10 audio clips) and makes 30-second shots with sound in one go. It is also the most expensive of the leading models.
- Cheapest way in
- Free starter generations in Dreamina; Basic from $15/month
- Cost per output
- About $0.57 per second at 1080p via API; $0.23 at 720p, $0.10 at 480p
- Max resolution
- 1080p (480p, 720p, 1080p)
- Max length
- 30s per generation, extendable twice
- Native audio
- Yes
- References
- Up to 30 images, 10 video clips, 10 audio clips
- Open weights
- No
- Where to use it
- Dreamina / CapCut, BytePlus ModelArk API, Runway, and other API hosts
Good at
- 30-second single generation, extendable twice
- Very large multimodal reference set for character, motion and audio control
- Consistently near the top on every arena, generation and editing alike
- Joint audio-video generation; also edits and extends existing footage
Watch out for
- Most expensive per second of the leading API models (about $0.57/s at 1080p)
- No 4K tier in 2.5 (Seedance 2.0 keeps 4K)
- Not a clear #1 on text-to-video: Wan 3.0, Utopai X and MiniMax H3 score higher on Artificial Analysis, and Google and BFL models score higher on Arena
Benchmarks Artificial Analysis Image to Video Arena: #1, Elo 1153 · Artificial Analysis Text to Video Arena (with audio): #4, Elo 1157 (behind Wan 3.0 1175, Utopai X 1166, MiniMax H3 1161) · Artificial Analysis Video Editing: #2, Elo 1152 · Arena (formerly LMArena) Text to Video: #7, 1474 +/-9 (720p; page dated 21 Sep 2026)
Prices ofox.ai, checked 1 November 2026
Seedance 2.5 website Work made with it on AI LIVE
2
Wan 3.0
Alibaba · Top-scoring text-to-video and editing at mid-range prices
$$
Wan 3.0 is #1 on Artificial Analysis text-to-video with audio and #1 on both public video-editing boards, and it makes 30-second clips. It is weaker on image-to-video (#9 on Artificial Analysis) and costs about a third of Seedance 2.5 at 1080p. Unlike earlier Wan models it is closed-weight.
- Cheapest way in
- Free 30-second quota on Alibaba Cloud Model Studio, then pay as you go
- Cost per output
- $0.20 per second at 1080p via API, $0.10 at 720p, $0.05 at 480p; Prime $0.068-0.28/s
- Max resolution
- 1080p
- Max length
- 30s per generation, with extension
- Native audio
- Yes
- References
- 10 images, 5 videos, 5 audio clips, plus PDFs, slides and web pages
- Open weights
- No
- Where to use it
- Alibaba Cloud Model Studio, Qwen Cloud, wan.video, Runway API
Good at
- Highest text-to-video Elo on Artificial Analysis
- 30-second single generations from text, image, video, audio and even documents
- Best-rated video editing on both arenas
- $0.20/s at 1080p on Alibaba's own API
Watch out for
- No open weights (the open Wan line stops at Wan 2.2)
- Only mid-table on Artificial Analysis image-to-video (#9)
- Access was beta-gated at launch; consumer wan.video is planned as members-only
Benchmarks Artificial Analysis Text to Video Arena (with audio): #1, Elo 1175 · Artificial Analysis Image to Video Arena: #9, Elo 1111 · Arena Text to Video: #6, 1476 +/-13 (page dated 21 Sep 2026) · Arena Image to Video: #3, 1480 +/-11 (page dated 21 Sep 2026)
Prices alibabacloud.com, checked 1 November 2026
Wan 3.0 website
3
MiniMax H3
MiniMax · Near-top quality for a fraction of the price, with downloadable weights
$
H3 is #1 on Arena image-to-video with the most votes of any leader there (57,112), and #3 and #4 on Artificial Analysis text- and image-to-video. Across the four generation boards its average rank matches Seedance 2.5 at roughly a quarter of the price per second. MiniMax also published weights under a community licence, though with commercial and regional limits.
- Cheapest way in
- Open weights free to run locally; API pay as you go from $0.025/s
- Cost per output
- $0.08 per second at 768p, $0.13 at 2K via API; Highspeed and Lite from $0.025/s
- Max resolution
- 2K via API; 768p for the open weights
- Max length
- 15s (4-15s)
- Native audio
- Yes
- References
- Up to 5 images; first/last frame; continuation; prompt editing
- Open weights
- Yes, MiniMax H3 Community License (restricted)
- Where to use it
- Hailuo app, MiniMax API, Hugging Face / ComfyUI, fal and other hosts
Good at
- Top-tier quality on every arena at $0.08-0.13 per second
- Native 2K output via API, with in-pass dialogue, effects and ambience
- Open weights (768p) on Hugging Face with ComfyUI support
- Cheap Highspeed and Lite variants from $0.025/s
Watch out for
- 15-second cap; longer scenes need stitching
- Up to 5 reference images, far fewer than Seedance 2.5
- Community licence: commercial use only under $20M annual revenue, and reports say local deployment is barred in the US, EU, UK and South Korea (check the licence)
Benchmarks Arena Image to Video: #1, 1495 +/-5 (page dated 21 Sep 2026) · Artificial Analysis Text to Video Arena (with audio): #3, Elo 1161 (768p) · Artificial Analysis Image to Video Arena: #4, Elo 1128 (768p) · Arena Text to Video: #8, 1460 +/-9 (page dated 21 Sep 2026)
Prices platform.minimax.io, checked 1 November 2026
MiniMax H3 website Work made with it on AI LIVE
4
Gemini Omni Flash 1.1
Google · Conversational generate-then-edit workflows inside Google's tools
$
Google's Veo successor leads Arena text-to-video (#1 and #2 across its two versions) and is #2 on Arena image-to-video. Artificial Analysis rates it lower (#10 text-to-video, #14 image-to-video), so the two arenas disagree on it more than on any other leader. It is cheap at about $0.10/s but capped at 10-second 720p clips.
- Cheapest way in
- In the Gemini app and Flow; API pay as you go
- Cost per output
- About $0.10 per second of 720p video via the Gemini API
- Max resolution
- 720p native, free 1080p upscale
- Max length
- 10s (4, 6, 8 or 10s)
- Native audio
- Yes
- References
- Up to 5 reference images; video input for editing
- Open weights
- No
- Where to use it
- Gemini app, Flow, YouTube Shorts, Gemini API, Runway
Good at
- #1 on Arena text-to-video
- Multi-turn conversational editing of generated or uploaded video
- About $0.10/s via the Gemini API
- Built into the Gemini app, Flow and YouTube Shorts
Watch out for
- 10-second maximum, no extension
- 720p native (1080p upscale)
- Much weaker showing on Artificial Analysis than on Arena
Benchmarks Arena Text to Video: #1, 1516 +/-15 (Gemini Omni 1.1 Flash; page dated 21 Sep 2026) · Arena Image to Video: #2, 1488 +/-11 (page dated 21 Sep 2026) · Artificial Analysis Text to Video Arena (with audio): #10, Elo 1128 · Artificial Analysis Image to Video Arena: #14, Elo 1046
Prices ai.google.dev, checked 1 November 2026
Gemini Omni Flash 1.1 website
5
FLUX 3 Video
Black Forest Labs · Up to 4K clips with keyframes and simple per-second pricing
$$
BFL's first video model is #3 on Arena text-to-video (preliminary) and its FLUX 3 Max variant ties Seedance 2.5 on Artificial Analysis text-to-video (Elo 1157). It offers 20-second clips, keyframes and resolutions up to UHD/4K with audio included at no extra charge. Image-to-video is weaker (#11-12 on Artificial Analysis).
- Cheapest way in
- Pay-as-you-go, no subscription; draft HD from $0.06/s
- Cost per output
- Text/image-to-video per second: $0.17 HD, $0.29 FHD, $0.40 QHD, $0.80 UHD/4K; video-to-video $0.41-0.95/s
- Max resolution
- UHD/4K
- Max length
- 20s
- Native audio
- Yes
- References
- Image-to-video plus start, end and multiple keyframes
- Open weights
- No
- Where to use it
- BFL dashboard and API, fal, Replicate, Together, Cloudflare and more
Good at
- Up to 4K output and 20-second clips
- Audio included in the price; multilingual lip-synced dialogue
- Start, end and multiple ordered keyframes; continuation and video-to-video
- Cheap draft mode ($0.06/s) for previews
Watch out for
- Image-to-video lags the leaders
- No open weights for the video model
- Arena placement is still preliminary
Benchmarks Arena Text to Video: #3, 1493 +/-17, preliminary (page dated 21 Sep 2026) · Artificial Analysis Text to Video Arena (with audio): #5, Elo 1157 (FLUX 3 Max); FLUX 3 #8, Elo 1144 · Arena Image to Video: #9, 1449 +/-5 (page dated 21 Sep 2026) · Artificial Analysis Image to Video Arena: #11, Elo 1071 (FLUX 3)
Prices bfl.ai, checked 1 November 2026
FLUX 3 Video website
6
HeyGen Video 1
HeyGen · Low-cost, high-scoring clips with dialogue and sound
$
HeyGen's first general video model is #1 on Artificial Analysis's silent text-to-video board and #6 on both its with-audio text-to-video and image-to-video boards. Artificial Analysis lists it at $2.40 per minute, among the cheapest models in the top 10. It is new (late September/October 2026) and not yet on Arena.
- Cheapest way in
- HeyGen Creative Studio plans or the HeyGen API
- Cost per output
- About $0.04 per second ($2.40 per minute) on Artificial Analysis; 1080p and 2K cost 3x the 768p rate
- Max resolution
- 2K (480p, 768p, 1080p, 2K)
- Max length
- 30s for text/image-to-video (turbo or quality); 15s with references
- Native audio
- Yes
- References
- Reference image and reference video inputs
- Open weights
- No
- Where to use it
- HeyGen Creative Studio, HeyGen API, OpenRouter, Runware, ComfyUI
Good at
- Top-10 quality on Artificial Analysis at a very low listed price
- Text-, image- and reference-to-video with stereo audio in the same pass
- Up to 30s for text/image-to-video in some enhancement modes
- Available in HeyGen Creative Studio and via OpenRouter, Runware and ComfyUI
Watch out for
- Official per-second rates not published in HeyGen's docs
- Very new, little independent creator testing yet
- Not yet rated on Arena
Benchmarks Artificial Analysis Text to Video Arena (silent): #1, Elo 1143 · Artificial Analysis Text to Video Arena (with audio): #6, Elo 1156 · Artificial Analysis Image to Video Arena: #6, Elo 1120
Prices artificialanalysis.ai, checked 1 November 2026
HeyGen Video 1 website
7
Grok Imagine Video 1.5
xAI · Fast, cheap iteration in the Grok app or API
$
It is #4 on Arena text-to-video (preliminary) and #8 on Arena image-to-video with more than 137,000 votes, but only #13 on both Artificial Analysis boards. xAI lists it at $0.08 per second, and a Lite version costs $0.02 per second.
- Cheapest way in
- In the Grok app (grok.com/imagine) and via the xAI API
- Cost per output
- $0.08 per second via the xAI API; Lite $0.02 per second
- Max resolution
- Up to 1080p (per Runway; xAI does not state it)
- Max length
- 15s (API accepts 1-15s)
- Native audio
- Yes (ranked on the with-audio arena)
- References
- Image-to-video and reference images
- Open weights
- No
- Where to use it
- grok.com/imagine, Grok apps, xAI API, Runway
Good at
- Strong Arena text-to-video placement
- Low official API price; Lite tier at $0.02/s
- Available in grok.com/imagine and mobile apps as well as the API
Watch out for
- 15-second cap
- Mid-table on Artificial Analysis
- Resolution and per-resolution pricing are inconsistently documented (xAI lists a flat $0.08/s; Runway resells at $0.10-0.29/s by resolution; Artificial Analysis lists $15/min)
Benchmarks Arena Text to Video: #4, 1492 +/-18, preliminary (page dated 21 Sep 2026) · Arena Image to Video: #8, 1456 +/-4 (720p; page dated 21 Sep 2026) · Artificial Analysis Text to Video Arena (with audio): #13, Elo 1059
Prices docs.x.ai, checked 1 November 2026
Grok Imagine Video 1.5 website Work made with it on AI LIVE
8
Utopai X
Utopai Studios · Film-style generation inside a production platform, at a high cost
$$$$
Utopai X is #2 on both Artificial Analysis text-to-video (Elo 1166) and image-to-video (Elo 1146), so on quality alone it belongs near the top. It ranks lower here because it has no public API, is not on Arena, and is only usable through the paid PAI platform, where 1080p is reported to cost 93 credits per second.
- Cheapest way in
- PAI Basic $15/month (1,000 credits); higher plans from $49/month
- Cost per output
- Reported at about $1.40 per second at 1080p on the Basic plan (93 credits per second)
- Max resolution
- 1080p; 4K export on Plus and above
- Max length
- Not published
- Native audio
- Not confirmed by Utopai
- Open weights
- No
- Where to use it
- pai.utopaistudios.com only
Good at
- Second-highest Elo on both Artificial Analysis generation boards
- Built into PAI, a production platform with multi-shot sequencing and story tools
Watch out for
- No public API; Artificial Analysis lists 'No API available'
- No free tier reported; on the $15 Basic plan, 1,000 credits cover roughly 10 seconds of 1080p
- Few published specs; Chinese reports say it is post-trained from MiniMax H3 (unconfirmed)
Benchmarks Artificial Analysis Text to Video Arena (with audio): #2, Elo 1166 · Artificial Analysis Image to Video Arena: #2, Elo 1146
Prices layer3labs.io, checked 1 November 2026
Utopai X website
9
Kling 4.0
Kuaishou · Creators already on Kling who want 30-second takes and big motion
$$
Kling 4.0 launched on 15 October 2026 with 30-second native clips and up to 15 reference assets and 10 keyframes, but it has no arena score yet and no public API pricing. The benchmarked Kling 3.0 now sits around #18-19 on Artificial Analysis (it is the 1000-Elo anchor) and #19 on Arena image-to-video, well behind the leaders. It ranks here on capability and access rather than measured quality.
- Cheapest way in
- Free plan without monthly credits; Standard $6.99 first month, then $8.80
- Cost per output
- Kling 3.0: 30 credits/s at 4K, or $0.12-0.17/s via Runway API; Kling 4.0 rates unpublished
- Max resolution
- Kling 3.0: native 4K; Kling 4.0: 4K/HDR reported, unconfirmed
- Max length
- Kling 4.0: 30s; Kling 3.0: 15s
- Native audio
- Yes
- References
- Kling 4.0: up to 15 reference assets and 10 keyframes
- Open weights
- No
- Where to use it
- kling.ai web and app; Kling 3.0 via Kling API, Runway, fal and others
Good at
- 30-second single-pass generation in 4.0
- Omni Reference: up to 15 reference assets and 10 keyframes
- Kling 3.0 offers native 4K and multi-shot; mature web app with cheap entry plan
Watch out for
- Kling 4.0 is unbenchmarked; Kling 3.0 trails the 2026 leaders
- Kling 4.0 API and credit costs not published; one report says membership plans do not yet cover 4.0
- 4K, HDR and stereo audio claims for 4.0 are reported but not confirmed by Kling
Benchmarks Artificial Analysis Text to Video Arena (with audio): Kling 3.0 Omni 1080p Pro #18, Elo 1023; Kling 3.0 1080p Pro #19, Elo 1000 (anchor); Kling 4.0 not listed · Arena Image to Video: Kling 3.0 Pro #19, 1354 +/-6 (page dated 21 Sep 2026)
Prices kling.ai, checked 1 November 2026
Kling 4.0 website Work made with it on AI LIVE
10
LTX-2.5
Lightricks · Running a capable audio-video model locally with open weights
$
LTX-2.5 is the strongest openly downloadable model with fewer strings than MiniMax H3: free commercial use under $10M revenue, native audio, and up to 4K output. Its quality is well behind the leaders (#25-26 on Artificial Analysis text-to-video, Elo about 945), so it is a pick for control, privacy and cost rather than polish.
- Cheapest way in
- Free: open weights, or LTX Studio free tier (800 one-time credits)
- Cost per output
- API reported at about $0.09 per second at 720p, $0.15 at 1080p, $0.37 at 4K
- Max resolution
- 4K (Fast); 1080p (Pro)
- Max length
- 20s at 720p/1080p; 10s at 1440p/4K
- Native audio
- Yes
- Open weights
- Yes, community licence (free commercial use under $10M revenue)
- Where to use it
- Hugging Face, ComfyUI, LTX Studio, LTX API
Good at
- Open weights (22B) on Hugging Face; runs in ComfyUI and the LTX desktop editor
- Native audio and up to 4K (Fast) output
- Free LTX Studio tier with 800 one-time credits
Watch out for
- Clearly behind closed models on every arena
- 4K clips limited to 10 seconds
- Paid licence needed above $10M annual revenue
Benchmarks Artificial Analysis Text to Video Arena (with audio): LTX-2.5 Fast #25, Elo 947; LTX-2.5 Pro #26, Elo 944 · Arena Text to Video: LTX-2 19B (older version) #41, 1154 +/-8; LTX-2.5 not listed
Prices kie.ai, checked 1 November 2026
LTX-2.5 website Work made with it on AI LIVE
