← All models

Doubao-Seed-2.1-Turbo

Published Publisher ByteDancedoubao-seed-2-1-turbo-260628

Positioning and capabilities

Doubao-Seed-2.1-Turbo is a deep-reasoning model positioned as a lower-cost, lower-latency Seed 2.1 option for production at scale. Its documented capabilities include text generation, multimodal understanding, GUI tasks, tool calling and structured output. ByteDance describes the Seed 2.1 series as improved in agent workflows, coding delivery and visual and video understanding. (docs.volcengine.com)

Uses

Suitable for high-volume coding assistance, agent workflows, multimodal understanding and enterprise deployments where cost, throughput and batch calls matter. (docs.volcengine.com)

Limits

Volcengine lists a 256K-token context window, maximum input of 256K tokens and maximum answer of 256K tokens, with a 4K-token default answer limit. Its listed limits are 500 requests per minute and 1,000,000 tokens per minute; these are not guaranteed and can depend on platform load and calling method. Outputs should be checked before use in consequential workflows. (docs.volcengine.com)

Release date

The official Seed 2.1 series announcement is dated 2026-06-23 and names Turbo. Volcengine's release record lists the exact model ID under June 2026; it does not establish a separate exact release date for that version. (seed.bytedance.com)

Pricing

Checked 2026-11-16. Prices below are official list prices for different platforms and service tiers, not currency conversions or interchangeable packages. Actual availability and billing should be confirmed in the relevant console.

  • Mainland China — Volcengine ModelArk, standard online inference: for input lengths of 0–256K tokens, non-audio input is CNY 3.00 per million tokens, output is CNY 15.00 per million tokens, non-audio cache hits are CNY 0.60 per million tokens, and cache storage is CNY 0.017 per million tokens per hour. The official billing page identifies the exact doubao-seed-2-1-turbo-260628 version. Official pricing source. (docs.volcengine.com)
  • Mainland China — other online tiers: for the same 0–256K input-length tier, low-priority inference lists non-audio input at CNY 1.50, output at CNY 7.50, and non-audio cache hits at CNY 0.60 per million tokens. Low-latency inference lists input at CNY 6.00, output at CNY 30.00, and cache hits at CNY 1.20 per million tokens. These are separate service tiers. Official pricing source. (docs.volcengine.com)
  • International — BytePlus ModelArk: BytePlus lists the related Seed 2.1 Turbo offering under the distinct ID dola-seed-2-1-turbo-260628. Its standard pay-as-you-go price is USD 0.50 per million input tokens and USD 2.50 per million output tokens. The BytePlus pricing documentation also lists USD 0.10 per million cached input tokens and USD 0.0083 per million cache-storage tokens per hour, for the 0–256K input-length tier. The international ID and platform differ from the requested mainland ID; identical deployment or behavior is not independently established by these listings. Official international pricing; pricing documentation. (byteplus.com)

Official sources