Doubao LLM API
Doubao LLM API
Doubao is ByteDance's self-developed LLM series, served via Volcengine Ark — Pro/Lite tiers covering dialog, long-context, vision, and embeddings.
Aggressively competitive pricing (initiator of the 2024 Chinese LLM price war)
No open-source release (unlike Qwen / Baichuan)
Sign up at volcengine.com → enable Doubao models in the Volcengine Ark console → create an inference endpoint (endpoint_id) → POST /api/v3/chat/completions (OpenAI-compatible).
Uptime · 30-day window
About this API
Doubao is ByteDance's self-developed LLM series, commercialized in 2024 and offered via the Volcengine Ark platform. The model matrix spans general dialog (Doubao-pro / Doubao-lite), 128K long-context (Doubao-pro-128k), vision (Doubao-vision), embeddings (Doubao-embedding), function calling, and image generation (Doubao-Seedream). Doubao is famous for initiating the 2024 Chinese LLM 'price war', dramatically lowering API prices and forcing Qwen / Zhipu / Baichuan to follow. Trained by ByteDance's internal AI Lab and battle-tested in the Doubao app, Douyin search, and CapCut AI. API access uses an 'inference endpoint' (endpoint_id) pattern — each model version maps to an endpoint for easy version switching. A natural fit for mainland-China developers building cost-sensitive Chinese LLM apps, especially enterprise scenarios needing low latency and compliance.
What you can build
- 1Chinese LLM apps (very low cost; battle-tested at ByteDance's own scale)
- 2Long-context processing (Doubao-pro-128k)
- 3Visual understanding (Doubao-vision)
- 4Integration scenarios within the Douyin / Toutiao ecosystem
Strengths & limitations
Strengths
- Aggressively competitive pricing (initiator of the 2024 Chinese LLM price war)
- Battle-tested at ByteDance scale (Doubao app, Douyin search, etc.)
- Enterprise-grade SLA and compliance support via Volcengine
Limitations
- No open-source release (unlike Qwen / Baichuan)
- Overseas developers need a Volcengine account and verification
- Some advanced models require enterprise vetting
Official quickstart
Read the official quickstart at www.volcengine.com.
Getting started
Sign up at volcengine.com → enable Doubao models in the Volcengine Ark console → create an inference endpoint (endpoint_id) → POST /api/v3/chat/completions (OpenAI-compatible).
FAQ
Doubao vs. Qwen — which to pick?+
Doubao is slightly cheaper; capability is comparable. Choose Doubao for ByteDance-ecosystem scenarios (Douyin etc.), Qwen for open-source breadth and multilingual reach.
Why is an endpoint_id required before calling?+
Volcengine Ark abstracts model versions behind endpoints, enabling gradual rollout — an externalization of ByteDance's internal ML-engineering practice, making multi-model multi-version management cleaner.
Technical details
- Auth type
- api_key
- Pricing
- paid
- Rate limit
- Per-endpoint and per-model; enterprise verification scales limits.
- Free tier quota
- Doubao Pro / Lite tiers each include 500K free tokens valid 30 days on signup. Vision and image endpoints carry separate quotas.
- Protocols
- REST
- SDKs
- python, javascript, java, go
- Response time
- 1343 ms
- Last health check
- 6/29/2026, 6:32:37 AM