October 2026 edition
AI model leaderboard: the best AI video and image generators, tested
We gave 12 video models and 12 image models on PopNow the same prompts and logged what every run cost, how long it took and how the result looked. No sponsored slots, no cherry-picking: one run per model per prompt, all 72 results shown below.
Tested Oct 17–19, 2026 · next update November 19, 2026
- Cheapest 5-second videoMiniMax H3185 credits · $0.19
- Fastest video modelMiniMax H329 s on average
- Highest-rated videoSeedance 2.0 Mini, Seedance 2.0, and Seedance 2.54.7 / 5 overall
- Cheapest imageZ-Image10 credits · $0.01
- Highest-rated imageGPT Image 2 and Nano Banana Pro5.0 / 5 overall
The leaderboard
Sort by what matters to you. Price is for one 5-second clip or one image at the settings we tested; time is from submit to finished file.
| # | Model | Output | Actions | ||||||
|---|---|---|---|---|---|---|---|---|---|
| 1 |  Seedance 2.0 MiniByteDance | 265 credits$0.27 per 5-second clip$0.053 per second | 104 s102–105 s | 720p1280×720 · 5 sSound | 4.5 | 5.0 | 4.5 | 4.7 | Try |
| 2 |  Seedance 2.0ByteDance | 760 credits$0.76 per 5-second clip$0.15 per second | 258 s257–259 s | 720p1280×720 · 5 sSound | 4.5 | 5.0 | 4.5 | 4.7 | Try |
| 3 |  Seedance 2.5ByteDance | 1,387 credits$1.39 per 5-second clip$0.28 per second | 236 s192–280 s | 720p1280×720 · 5 sSound | 5.0 | 4.5 | 4.5 | 4.7 | Try |
| 4 |  Wan 3.0Alibaba | 450 credits$0.45 per 5-second clip$0.09 per second | 99 s98–100 s | 720p1280×720 · 5 sSound | 4.5 | 4.5 | 4.0 | 4.3 | Try |
| 5 |  Seedance 2.0 FastByteDance | 484 credits$0.48 per 5-second clip$0.097 per second | 157 s102–211 s | 720p1280×720 · 5 sSound | 4.5 | 4.5 | 4.0 | 4.3 | Try |
| 6 |  HappyHorse 1.1Alibaba | 700 credits$0.70 per 5-second clip$0.14 per second | 101 s | 720p1280×720 · 5 sSound | 4.0 | 5.0 | 4.0 | 4.3 | Try |
| 7 |  Kling 4.0Kuaishou | 1,080 credits$1.08 per 5-second clip$0.22 per second | 808 s806–810 s | 720p1280×720 · 5 sSound | 4.5 | 4.5 | 4.0 | 4.3 | Try |
| 8 |  MiniMax H3MiniMax | 185 credits$0.19 per 5-second clip$0.037 per second | 29 s27–30 s | 768p1280×732 · 5 sSound | 4.0 | 4.5 | 4.0 | 4.2 | Try |
| 9 |  Kling 3.0Kuaishou | 504 credits$0.50 per 5-second clip$0.10 per second | 175 s101–248 s | 720p1280×720 · 5 sSound | 4.0 | 4.0 | 4.0 | 4.0 | Try |
| 10 |  Kling O3Kuaishou | 448 credits$0.45 per 5-second clip$0.09 per second | 239 s218–260 s | 720p1280×720 · 5 sSound | 3.5 | 4.5 | 3.5 | 3.8 | Try |
| 11 |  HunyuanVideo 1.5Tencent | 240 credits$0.24 per 5-second clip$0.048 per second | 211 s208–214 s | 720p1280×720 · 5 sNo sound | 3.0 | 2.5 | 3.5 | 3.0 | Try |
| 12 |  LTX 2.5Lightricks | 450 credits$0.45 per 5-second clip$0.09 per second | 47 s34–59 s | 720p1280×704 · 5 sSound | 2.5 | 3.0 | 2.5 | 2.7 | Try |
| # | Model | Output | Actions | ||||||
|---|---|---|---|---|---|---|---|---|---|
| 1 |  GPT Image 2OpenAI | 43 credits$0.043 per square 1K image | 62 s60–65 s | 1K1254×1254 | 5.0 | 5.0 | 5.0 | 5.0 | Try |
| 2 |  Nano Banana ProGoogle | 108 credits$0.11 per square 1K image | 30 s27–34 s | 1K1024×1024 | 5.0 | 5.0 | 5.0 | 5.0 | Try |
| 3 |  Seedream 5.0 FlashByteDance | 29 credits$0.029 per square 1K image | 28 s21–32 s | 1K1024×1024 | 4.8 | 5.0 | 5.0 | 4.9 | Try |
| 4 |  GPT Image 2.5 SunburstOpenAI | 43 credits$0.043 per square 1K image | 71 s62–85 s | 1K1254×1254 | 5.0 | 5.0 | 4.8 | 4.9 | Try |
| 5 |  Qwen Image 3.0Alibaba | 30 credits$0.03 per square 1K image | 76 s59–82 s | 1K1024×1024 | 4.3 | 5.0 | 5.0 | 4.8 | Try |
| 6 |  Krea 3Krea | 36 credits$0.036 per square 1K image | 14 s13–14 s | 1024×1024 | 4.3 | 5.0 | 5.0 | 4.8 | Try |
| 7 |  FLUX.2 MaxBlack Forest Labs | 70 credits$0.07 per square 1K image | 48 s35–81 s | 1K1024×1024 | 4.3 | 4.8 | 5.0 | 4.7 | Try |
| 8 |  Seedream 5.0 ProByteDance | 36 credits$0.036 per square 1K image | 57 s43–79 s | 1K1024×1024 | 4.0 | 4.8 | 5.0 | 4.6 | Try |
| 9 |  Z-ImageAlibaba | 10 credits$0.01 per square 1K image | 12 s11–12 s | 1024×1024 | 4.0 | 4.3 | 5.0 | 4.4 | Try |
| 10 |  Wan 2.7Alibaba | 30 credits$0.03 per square 1K image | 31 s25–35 s | 1K1024×1024 | 3.8 | 4.8 | 4.8 | 4.4 | Try |
| 11 |  Nano Banana 2.1Google | 48 credits$0.048 per square 1K image | 31 s28–37 s | 1K1024×1024 | 4.0 | 4.3 | 5.0 | 4.4 | Try |
| 12 |  FLUX.2 ProBlack Forest Labs | 30 credits$0.03 per square 1K image | 23 s19–27 s | 1K1024×1024 | 4.0 | 4.0 | 4.8 | 4.3 | Try |
Tick two models to compare them side by side.
Prices are today’s catalog prices for these exact settings: the same number the Generate button shows. 1 credit = $0.001 at list price. Times include any queue and change with load, so treat them as a guide.
Key numbers this month
- A 5-second clip cost between 185 credits and 1,387 credits at each model’s default resolution: $0.037 to $0.28 per second of video.
- The fastest video model, MiniMax H3, finished in 29 seconds on average; the slowest, Kling 4.0, took 808 seconds.
- 11 of 12 video models returned a soundtrack with the clip; 1 returned silent video.
- Images cost between 10 credits and 108 credits each at 1K and took 12 to 76 seconds on average.
- On the poster-text test, the top score went to GPT Image 2.5 Sunburst, GPT Image 2, and Nano Banana Pro.
Every result, same prompt
Pick a prompt to see what each model made from it. Results are shown as they came out; we never re-rolled for a better result.
Video models
Golden hour on an old wooden pier by the sea. A Black street musician in his forties, wearing a gray knit beanie, sits on a crate playing acoustic guitar with his case open in front of him. A young woman with red hair and a denim jacket glides past on roller skates, spins once, and drops a coin into the guitar case; he grins and nods to her. Seagulls drift overhead and the water glitters behind them. Camera: a smooth tracking shot that follows her, then settles on him in a medium shot. Sound: a warm fingerpicked guitar melody, skate wheels rolling on wooden planks, waves, gulls, and the clink of the coin.
Hover or tap a clip to play it.
- Seedance 2.5Try this prompt
- Seedance 2.0 MiniTry this prompt
- Wan 3.0Try this prompt
- HappyHorse 1.1Try this prompt
- Kling O3Try this prompt
- Seedance 2.0 FastTry this prompt
- Seedance 2.0First attempt was stopped by the model’s copyright filter (it flagged the generated music). Shown: the second attempt with the same settings.Try this prompt
- MiniMax H3Try this prompt
- Kling 3.0Try this prompt
- Kling 4.0Try this prompt
- LTX 2.5Try this prompt
- HunyuanVideo 1.5Try this prompt
A golden retriever puppy chases a giant wobbling orange jelly cube across a sunlit park lawn. The cube bounces, squashes on landing and splits into three smaller glossy cubes that scatter in different directions; the puppy skids to a stop, tilts its head and pounces on the nearest one. Low tracking camera at grass height, shallow depth of field, warm afternoon light. Sound: excited barks, squishy bounces, birdsong.
Hover or tap a clip to play it.
- Seedance 2.0Try this prompt
- Seedance 2.0 MiniTry this prompt
- Kling 4.0Try this prompt
- MiniMax H3Try this prompt
- Seedance 2.0 FastTry this prompt
- Seedance 2.5Try this prompt
- Wan 3.0Try this prompt
- Kling 3.0Try this prompt
- HappyHorse 1.1Try this prompt
- Kling O3Try this prompt
- HunyuanVideo 1.5Try this prompt
- LTX 2.5Try this prompt
Image models
Retro sci-fi photograph with deadpan humor: a couple in their seventies in matching mint tracksuits have a picnic on a pale grey lunar landscape, checked blanket, a thermos and a cherry pie, a single giant iridescent soap bubble floating beside them, Earth rising huge on the horizon behind them; crisp hard light, 1970s film color, photorealistic, no text.
GPT Image 2.5 SunburstTry this prompt
GPT Image 2Try this prompt
Seedream 5.0 FlashTry this prompt
Qwen Image 3.0Try this prompt
Nano Banana ProTry this prompt
Krea 3Try this prompt
FLUX.2 MaxTry this prompt
Z-ImageTry this prompt
FLUX.2 ProTry this prompt
Wan 2.7Try this prompt
Seedream 5.0 ProTry this prompt
Nano Banana 2.1Try this prompt
Studio product photo for an online shop: a matte sage-green ceramic coffee mug with a small embossed label that reads "MORNING", standing on a light oak table next to a folded linen napkin and three roasted coffee beans. Soft window light from the left, a gentle natural shadow, clean warm-white background, sharp focus on the glaze texture and the rim, true-to-life colors, no logos, no brand names, no other text.
Seedream 5.0 FlashTry this prompt
GPT Image 2.5 SunburstTry this prompt
GPT Image 2Try this prompt
FLUX.2 MaxTry this prompt
Nano Banana ProTry this prompt
FLUX.2 ProTry this prompt
Qwen Image 3.0Try this prompt
Wan 2.7Try this prompt
Seedream 5.0 ProTry this prompt
Krea 3Try this prompt
Nano Banana 2.1Try this prompt
Z-ImageTry this prompt
Hyperreal wide photograph of a candy-colored retro arcade at night: rows of pastel arcade cabinets in lilac, mint and tangerine, a glowing neon sign on the back wall that reads "POP ARCADE" in rounded letters, a tall woman in her thirties with dark curly hair and a cobalt satin bomber jacket leans on a claw machine filled with glossy jelly toys, haze and neon reflections on a glossy floor, cinematic composition, crisp readable sign, no other text, no logos.
GPT Image 2Try this prompt
Seedream 5.0 FlashTry this prompt
Krea 3Try this prompt
Nano Banana ProTry this prompt
GPT Image 2.5 SunburstTry this prompt
Qwen Image 3.0Try this prompt
Seedream 5.0 ProTry this prompt
Z-ImageTry this prompt
FLUX.2 ProTry this prompt
FLUX.2 MaxTry this prompt
Wan 2.7Try this prompt
Nano Banana 2.1Try this prompt
Photograph of a bold typographic event poster pasted on a sunlit concrete wall: giant chunky rounded letters "JELLY FEST" in tangerine and cobalt, a subtitle "SUMMER PARADE — 08.08" in a clean grotesk typeface, layered with cut-out stickers of wobbling jelly cubes, iridescent soap bubbles and cherries, halftone riso-print texture, lemon-yellow and lilac color blocks; harsh afternoon sun casts a diagonal shadow across the poster. Editorial graphic design, crisp legible text, no logos, no brand names.
GPT Image 2.5 SunburstTry this prompt
GPT Image 2Try this prompt
Nano Banana ProTry this prompt
Z-ImageTry this prompt
Seedream 5.0 FlashTry this prompt
Qwen Image 3.0Try this prompt
Wan 2.7Try this prompt
Seedream 5.0 ProTry this prompt
Krea 3Try this prompt
Nano Banana 2.1Try this prompt
FLUX.2 MaxTry this prompt
FLUX.2 ProTry this prompt
What we saw, model by model
One line per model from the editor who scored the runs. These are notes on this month’s results, not a verdict on the model.
Video models
- Seedance 2.0 Mini
Nearly matched the full Seedance 2.0 on both prompts for about a third of the price, and finished in under two minutes.
- Seedance 2.0
Split the jelly cube into exactly three pieces in mid-air, as written. Its first pier attempt was stopped by the model’s own music-copyright filter; the retry went through.
- Seedance 2.5
Top marks on the people-and-sound prompt: the spin, the coin and the closing medium shot all landed. The priciest clip in this test, and it took 3 to 5 minutes.
- Wan 3.0
Lively, sharp clips with sound: the gulls, the spin and the coin drop were all there. Both clips finished in under two minutes.
- Seedance 2.0 Fast
Followed both shot lists closely for about two thirds of the price of Seedance 2.0. The cube split into four pieces instead of three.
- HappyHorse 1.1
The most cinematic camera on the pier: it followed the skates, then settled on the musician. On the motion test the cube changed size between shots.
- Kling 4.0
Hit every beat of the motion prompt (squash, three cubes, head tilt, pounce) with crisp detail, but each clip took about 13 minutes, the slowest in the test.
- MiniMax H3
By far the fastest here: both clips were ready in about 30 seconds, with sound, at 768p. On the pier it skipped the coin drop.
- Kling 3.0
Smooth motion and a clear spin, but the camera stayed wide instead of settling on the musician, and the cube broke into more than three pieces.
- Kling O3
Tracked the skater nicely on the pier. On the motion test the cube melted into a blob before splitting, and the puppy’s shape wobbled.
- HunyuanVideo 1.5
No soundtrack, and on the pier it held one wide shot and skipped the spin and the coin. The cube split roughly on the motion test.
- LTX 2.5
Quick (under a minute) and with sound, but the colors were heavily oversaturated and the cube never split as asked.
Image models
- GPT Image 2
Full marks on all four prompts at its default medium quality. About a minute per image.
- Nano Banana Pro
Top marks on all four prompts: crisp poster text, warm 1970s film color on the moon picnic and a clean product shot. The priciest image model here.
- Seedream 5.0 Flash
Fast, inexpensive and nearly flawless: within a tenth of a point of the top score on our four prompts.
- GPT Image 2.5 Sunburst
Rich detail and the strongest 1970s mood on the picnic. About a minute per image, and a little stray text crept into the arcade’s floor reflection.
- Qwen Image 3.0
Clean, accurate results on every prompt, but one of the slowest image models here at more than a minute per image.
- Krea 3
One of the two fastest image models (about 14 seconds), with correct poster text and a vivid arcade.
- FLUX.2 Max
Natural light and a convincing product shot. The arcade came out darker and moodier than the candy colors we asked for.
- Seedream 5.0 Pro
Realistic people and correct poster text; on the picnic the Earth was barely in frame. Noticeably slower than Seedream 5.0 Flash.
- Z-Image
The cheapest and fastest image model in this test (about 12 seconds). Fine for drafts; the mug label came out printed rather than embossed.
- Wan 2.7
Got the text and objects right, but the people looked more like CGI than photos, and the arcade sign ran off the edge.
- Nano Banana 2.1
Clean poster text at a low price. It hid half of the neon sign behind the woman’s head and drew the Earth small instead of huge on the horizon.
- FLUX.2 Pro
Quick and inexpensive, but it printed the poster’s date line twice and added small logos to the tracksuits.
How we test
- Same prompts for every model: four for images (people, product, a scene with a neon sign, poster text) and two for video (people with sound, motion and physics). The full prompts are above, word for word.
- Default settings. Video: 5 seconds, 16:9, each model’s default resolution, sound on where the model offers it. Images: 1K where there is a size choice, the same aspect ratio for every model, default quality.
- One run per model per prompt, run through PopNow on Oct 17–19, 2026. We show the first result, unless the model refused the request; then we note it and show the second attempt.
- Price is what the Generate button shows for those exact settings, in credits and in dollars (1 credit = $0.001). Time runs from the moment the job is submitted to the moment the file is ready, including any queue.
- Every result is scored 1–5 on three things (below) by one editor looking at it full size. Scores are opinions, not lab measurements: a one-point gap is within the noise.
How we score
- Visual quality
- Detail, light, texture and realism. 5: you could post it as is. 3: usable after a retry or a touch-up. 1: unusable.
- Prompt following
- Did it include what we asked for: the people, objects, text, action and camera move? 5: everything. 4: one small miss. 3: a main element missing. 1: ignored the prompt.
- Consistency
- Images: hands, faces, text and objects hold up when you zoom in. Video: people and objects keep their shape and identity for the whole clip and move believably. 5: no glitches. 3: visible glitches. 1: falls apart.
We record whether a clip has sound, but this edition does not score the soundtrack itself.
Not in this edition: models that only animate a photo (such as Vidu Q4), because they can’t take a text-only prompt.
We rerun the same prompts every month, add models that launched in the meantime and refresh prices. Next update: November 19, 2026.
Using these numbers? Please credit “PopNow AI Model Leaderboard (October 2026)” and link to this page.
Download the data (CSV)Head-to-head comparisons
Two models, side by side, with the same prompt run on both.
- Kling 3.0 vs Seedance 2.5
- Kling 3.0 vs Wan 3.0
- Seedance 2.5 vs Wan 3.0
- Kling 3.0 vs Kling O3
- Seedance 2.0 vs Seedance 2.5
- Kling 3.0 vs MiniMax H3
- Nano Banana 2.1 vs GPT Image 2.5 Sunburst
- Nano Banana 2.1 vs Nano Banana Pro
- Seedream 5.0 Pro vs Nano Banana 2.1
- FLUX.2 Max vs Nano Banana Pro
- GPT Image 2 vs Nano Banana 2
- GPT Image 2 vs Nano Banana Pro
- Seedance 2.0 Mini vs Seedance 2.0 Fast
- Seedance 2.0 Fast vs Seedance 2.0
- HappyHorse 1.1 vs Seedance 2.0
- Seedance 2.0 vs Kling 3.0
- All comparisons
- Sora alternatives
- All AI models
Questions about the leaderboard
What is the best AI video generator right now?
In our October 2026 test, the highest overall score (4.7 out of 5) went to Seedance 2.0 Mini, Seedance 2.0, and Seedance 2.5. The best pick depends on the job: MiniMax H3 was the cheapest at 185 credits ($0.19) for a 5-second clip, and MiniMax H3 was the fastest. Sort the table by price, speed or score to choose for yours.
What is the cheapest AI video generator?
MiniMax H3: 185 credits ($0.19) for a 5-second clip at its default resolution, about $0.037 per second. The most expensive clip in this test cost 1,387 credits ($1.39).
How much does AI video cost per second?
Across the 12 video models we tested, a second of video cost between $0.037 and $0.28 at each model’s default resolution. Higher resolutions and longer clips cost more; each model page lists its full price table.
Which AI video generator is the fastest?
MiniMax H3 returned its clips in 29 seconds on average, from submit to download. The slowest model, Kling 4.0, averaged 808 seconds. Times include queueing and change with load.
Which AI image generator is best?
The highest overall image score (5.0 out of 5) went to GPT Image 2 and Nano Banana Pro. On the poster-text test, the top score went to GPT Image 2.5 Sunburst, GPT Image 2, and Nano Banana Pro. The cheapest image model was Z-Image, at 10 credits ($0.01) per square image.
How do you test, and can I use the data?
Every model gets the same prompts at its default settings, one run each. We log the price shown on the Generate button, the time from submit to finished file, and score each result 1–5 by eye. You can download the whole table as CSV; please credit “PopNow AI Model Leaderboard” and link to this page.
How often is the leaderboard updated?
Monthly. We rerun the same prompts, add new models and refresh prices. Next update: November 19, 2026.