Doubao
Doubao is ByteDance's own model family, covering text, vision, video (Seedance), image (Seedream), speech, simultaneous interpretation and embeddings, sold to developers through the Volcano Engine Ark platform. ByteDance advertises 180 trillion tokens a day flowing through it, and the same models reach non-China developers under Seed names on BytePlus ModelArk.
Pros
- One of the cheapest frontier-class families anywhere, with Doubao-Seed-2.1-turbo at CN¥3 per million input tokens and CN¥15 per million output, cached input dropping to CN¥0.6/M and entry tiers advertised from CN¥0.15/M input
- the international Seed line on BytePlus ModelArk goes lower still at $0.075/M input for seed-1-6-flash and halves again on batch inference. Text, vision, video, image, TTS, voice cloning and embeddings all sit behind a single OpenAI-compatible API with free trial quota on both consoles, so one integration covers the whole modality stack rather than stitching together four vendors.
Cons
- The Volcano Engine console, documentation and billing are Chinese-language and built around mainland accounts with real-name verification, which pushes outside developers to BytePlus ModelArk where the branding disappears — Doubao-Seed-2.0-pro becomes seed-2-0-pro — and the catalogue trails the Chinese one, with Doubao-Seed-Evolving and Seed-2.1-pro not yet mirrored. Model IDs carry date suffixes that rotate every few months, prices are adjusted with short notice, and there are no open weights to fall back on if a version is retired.
Cost
On Volcano Engine Ark: Doubao-Seed-2.1-turbo CN¥3/M input and CN¥15/M output (about $0.42 and $2.11), Doubao-Seed-2.1-pro and Doubao-Seed-Evolving CN¥6/M input and CN¥30/M output, cache-hit input CN¥1.2/M and cache storage CN¥0.017/M-tokens/hour; Seedream 5.0 images from CN¥0.22–0.30 each, speech synthesis CN¥5 per 10,000 characters, streaming speech recognition CN¥4.5/hour. Internationally on BytePlus ModelArk, billed in USD: seed-2-0-mini $0.10/M input and $0.40/M output, seed-2-0-pro $0.50/M and $3.00/M, seed-1-6-flash $0.075/M and $0.30/M, dola-seed-2-1-turbo $0.50/M and $2.50/M, with cache-hit input from $0.015/M and batch inference at roughly half list. Free trial quota is offered on both platforms.
Verdict
Best for teams shipping into Chinese-speaking markets or running high-volume video, speech and image generation alongside text, where the per-token gap against Western labs compounds fast; start on BytePlus ModelArk rather than the mainland console unless you already hold a Chinese business entity.
Visit Doubao