The 2026 video model lineup — what each one is actually good at — step 2 of 9

Veo 3.1, Omni, and Sora 2 — three APIs, three jobs

Omni, Veo 3.1, and Sora 2 are the closest thing to "GPT-4 class" video models in 2026. They are not the same product, even though casual coverage talks about them like they are.

Omni 1.1 Flash — the default Google call

Gemini Omni 1.1 Flash is the default video model on the Gemini API. The ID is gemini-omni-1.1-flash. You call it with client.interactions.create. Set video_config.task to text_to_video or image_to_video. Do not send Omni to generate_videos.

Agent Platform uses gemini-omni-1.1-flash-preview for the same model. The Gemini API GA ID has no -preview suffix. gemini-omni-flash-preview is a different, deprecated ID. It is removed 2026-09-30.

Omni is token-billed, roughly $0.10/sec at 720p (~$6/min raw). One-shot image-to-video is a valid call: you can hand it a frame and get motion without a second model in the stack. By default it also generates an audio track; you prompt the sound (music, energy, the room). That is not the same claim as Veo's speech / ambience / music pass — don't treat them as one product.

Veo 3.1 — cinematic + audio-native

Google's Veo family on a current endpoint is 3.1. veo-3.0-* and veo-2.0-* shut down on the Gemini API on 2026-06-30.

• Veo 3.1 Standard: 720p / 1080p. $0.40/sec with audio — $24/min raw. Speech with lip-sync. Ambient sound. Music. All generated in the same pass as the picture. No compositing step. Gemini preview ID veo-3.1-generate-preview. Agent Platform GA veo-3.1-generate-001.
• Veo 3.1 Fast: $0.10/sec at 720p with audio. veo-3.1-fast-generate-preview / veo-3.1-fast-generate-001.
• Veo 3.1 Lite: $0.05/sec at 720p with audio — $3/min raw. Shipped March 31, 2026. Quality sits below Standard; this is the cheap Veo. veo-3.1-lite-generate-preview / veo-3.1-lite-generate-001.

Call Veo with generate_videos or generateContent.

What Veo 3.1 does better than Omni:

• Speech, ambience, and music in one shot, on a dedicated video family. When the brief is "talking head with synchronized speech" and you want the Veo look, Standard is the default. Lite and Fast ship the same audio pass at the cheap and mid rungs — Standard is the look and the price, not the only native-audio ID.
• A cheap audio tier. Lite at $0.05/sec is the volume answer inside Google. Omni is ~$0.10/sec and also ships sound; it is not a silent default.

What it does NOT do better:

• Being the default Google call. That's Omni.
• First-frame stills. Those are gemini-3.1-flash-image (or gemini-3-pro-image) via generate_content. gemini-2.5-flash-image retires 2026-10-02.

Sora 2 — physics-first, removed 2026-09-24

OpenAI shipped the Sora 2 API at DevDay on October 6, 2025. The Videos API is removed 2026-09-24. Until then it still runs. Do not pick it as the production default.

• Sora 2 (base): 720p, $0.10 per second. Native synchronized audio.
• Sora 2 Pro: $0.30/sec at 720p, $0.50/sec at 1024p, $0.70/sec at 1080p. Same audio.

Jobs land in the 16-20 second range. A 60-second piece is three extends, not one call. The "$30 for 60s at 1024p" figure is 60 × $0.50 — raw seconds, not a single request. Moderation failures still get billed.

What Sora 2 still does well, through that date: character and physics consistency. A person's face is the same face from frame 1 to the end of the job. A poured glass of water behaves like water. Other models still burn obvious tells (fingers melting, faces shifting, fluids behaving like jello).

What it does NOT do better:

• Camera moves. You can prompt a dolly-in or a rack focus and you get something, but it's nondeterministic. Higgsfield (next step) exists because the base models are not yet good at this.
• Long durations in one shot. 16-20s jobs, then extend.
• Being the thing you put in a production router after 2026-09-24.

The "which one" question

Three-way split that works in practice:

Job	Use
Default Google video, including one-shot I2V	Omni (gemini-omni-1.1-flash via interactions.create)
Talking-head ad with VO, 30 sec	Veo 3.1 Standard (veo-3.1-generate-preview, $12.00)
Mass B-roll, 60 clips × 6 sec	Veo 3.1 Lite ($0.30/clip × 60 = $18)
Cheap 720p+audio, need Fast	Veo 3.1 Fast ($0.10/sec)
Physics demo while Videos API is up	Sora 2 — not the production default

When in doubt: Omni if you want the Google default; Veo 3.1 if you want the Veo look and the Lite/Fast/Standard price ladder; Sora 2 only if you know it is gone 2026-09-24.

First frames

If you do author a still before the video call, the live Google image ID is gemini-3.1-flash-image (or gemini-3-pro-image) via generate_content. Imagen 4 (imagen-4.0-*) on the Gemini API died 2026-08-17. A 1K still is ~$0.067; an 8-second Lite clip is $0.40. You stack the still because you want the frame, not because the still is the cheap part of the bill.