Skip to main content

OpenAIAlibabaVideo model matchup

Sora 2vs.Lumos-1 3B SFT 384p

Lumos-1 3B SFT 384p has no LMArena rating yet. Sora 2 has been discontinued.

Text to videoLMArena rating
1,343

Sora 2

Place 12 of 23

n/a

Lumos-1 3B SFT 384p

Longest clipSeconds per request
20 s

Sora 2

n/a

Lumos-1 3B SFT 384p

Choose Sora 2 for

No clear advantage

Choose Lumos-1 3B SFT 384p for

open weights

Same prompt, both models

The first clip from a single request each, showing the first five seconds.

What to look for: Anatomy, object consistency, camera tracking, rain, and reflections. Sound muted, as the test protocol requires.

Sora 2

Run details

fal.ai. fal-ai/sora-2/text-to-video. 16:9, one request, first output, requested 8 s, seed provider random, request {"prompt":"Generate exactly one continuous five-second 16:9 cinematic video. A fictional bicycle courier in a mustard-yellow raincoat rides a red city bicycle from left to right across a rain-wet Hamburg street at blue hour for the entire clip. The camera tracks her smoothly at waist height. Both hands remain visible on the handlebars, the bicycle geometry stays physically correct, and a transparent umbrella fixed to the bicycle moves naturally in the wind. Wet-street reflections and raindrops follow realistic physics. No cuts, no slow motion, no text, no logos, no subtitles, and no audio.","resolution":"720p","aspect_ratio":"16:9","duration":8}, configured price estimate $0.80.

Generated Aug 24, 2026. Source

Lumos-1 3B SFT 384p

No clip: this task has not yet been tested for the model.

Show the prompt given to both models

Generate exactly one continuous five-second 16:9 cinematic video. A fictional bicycle courier in a mustard-yellow raincoat rides a red city bicycle from left to right across a rain-wet Hamburg street at blue hour for the entire clip. The camera tracks her smoothly at waist height. Both hands remain visible on the handlebars, the bicycle geometry stays physically correct, and a transparent umbrella fixed to the bicycle moves naturally in the wind. Wet-street reflections and raindrops follow realistic physics. No cuts, no slow motion, no text, no logos, no subtitles, and no audio.

Where is the difference?

Text to videoNo rating for Lumos-1 3B SFT 384p
Longest clipNo clip length for Lumos-1 3B SFT 384p
ResolutionSora 2 higher720p to 384p
SoundSora 2 with sound

Bar length shows the gap: for ratings full width from 60 points, for price from four times as much, for clip length from 20 seconds. Solid means clearly ahead, hatched slightly ahead, a thin grey bar level.

All data side by side

The better value is bold. For ratings only when the gap is clear.

MetricSora 2Lumos-1 3B SFT 384p
Performance
Text to videoLMArena rating1,343Place 12 of 23n/a
Output
Resolution720p384p
Longest clip20 sn/a
Soundgenerated by the modelyesno
Access
Weightsproprietaryopen
Parametersn/a3B
Routesno longer offeredlocal
ReleasedSeptember 2025March 2026
StatusdiscontinuedDiscontinued: Sora 2 and Videos API since September 24, 2026available

Prices: public API price of the documented standard configuration; resolution and sound can change it

Sora 2 and Lumos-1 3B SFT 384p in the field

Every video model of our comparison with a rating, both models of this matchup highlighted.

Text to video

Rating from anonymous comparisons in the LMArena · higher is better

Gemini Omni 1.1 Flash1,516
Gemini Omni Flash1,513
FLUX.3 Video1,493
Grok Imagine Video 1.51,492
Dreamina Seedance 2.0 720p1,479
Dreamina Seedance 2.5 720p1,474
MiniMax H31,460
Muse Video1,456
Sora 2 Pro1,368
Veo 3.11,364
Veo 3.1 Fast1,362
Sora 21,343
Grok Imagine Video1,342
Wan 2.71,337
Wan 2.61,327
PixVerse V5.61,240
Runway Gen-4.51,225
Kling 2.6 Pro1,216
Hailuo 2.31,206
HunyuanVideo 1.51,169
LTX-2 19B1,154
Wan 2.2 A14B1,132
Pika 2.21,009

As of September 22, 2026 · Source: LMArena

Is there a better choice?

Best model for text to videoGemini Omni 1.1 Flash

Text to video 1,516, Place 1 of 23. $0.51 per 5 s clip.

Sora 2 vs. Gemini Omni 1.1 Flash

Sources and method

Show 4 sources and the method

Ratings come from the LMArena: people see two videos for the same prompt without knowing which model made them and pick the better one. Many such picks add up to a rating, higher is better. Text to video and image to video are separate lists and are never added together.

“Ahead” means one model stays clearly higher even with the rating’s uncertainty. “Slightly ahead” means its value is higher but the gap is not certain yet. Otherwise both count as level. Places count only the models of our comparison.

Prices and clip lengths come from the providers’ documentation. From a factor of 1.25 a model counts as cheaper or longer, below that as about the same.

Compare another matchup