Volcengine Ark Doubao-Seed-2.1-Turbo Model ID and Online/Batch Pricing Configuration

One-sentence takeaway

For budget-sensitive Agent routing with doubao-seed-2.1-turbo, clearly distinguish online from Batch pricing, and record the 256K input limit and the actual model snapshot in the evaluation configuration.

Use cases

• Suitable tasks: High-throughput, low-latency Agents; batch coding/extraction; and Pro/Turbo cost A/B testing and budget estimation.

• Unsuitable tasks: Treating the pricing table as a complete API schema, or using the Batch unit price to promise online latency.

• Applicable model version: Online model doubao-seed-2.1-turbo; it must not be mixed with doubao-seed-2.1-pro.

• Applicable clients, Agents, or APIs: Volcengine Ark online calls and Batch; confirm the endpoint, authentication, and regional availability in the current console.

• Recommended reasoning tier and parameters: Not publicly disclosed; record the reasoning toggle, maximum output, temperature, tools, and retry strategy as required fields.

Ready-to-use content

This template normalizes the official page into routing configuration. The field names are for internal use and are not the complete official API schema:

provider: volcengine-ark
model: doubao-seed-2.1-turbo
mode: online  # online or batch; do not use the same price budget
max_input_context: 256k
pricing:
  input_yuan_per_million_tokens: 3.00
  cached_input_yuan_per_million_tokens: 0.60
  output_yuan_per_million_tokens: 15.00
batch_pricing:
  input_yuan_per_million_tokens: 1.50
  cached_input_yuan_per_million_tokens: 0.60
  output_yuan_per_million_tokens: 7.50
record_for_each_run:
  endpoint_snapshot: required
  reasoning_mode: required
  prompt_tokens: required
  completion_tokens: required
  retries: required

Test/workflow steps

1. Verify the model ID, region, purchasable specifications, and endpoint in the Ark console; the official page states that the order result is the final basis for specifications and fees.

2. For online tasks, record input, cache hits, output, latency, and retries; for Batch tasks, separately record queue and completion times.

3. For long-context tasks, check whether the input exceeds the 256K range listed on the page, and record the chunking strategy.

4. After a pricing or model snapshot update, retain the collection date and rerun the same task set.

Original evidence and data

• The official page lists the online input length for doubao-seed-2.1-turbo as [0,256]K, with input priced at 3.00 yuan per million tokens, cache hits at 0.60 yuan per million tokens, and output at 15.00 yuan per million tokens.

• The same page lists Turbo Batch input at 1.50 yuan per million tokens, cache hits at 0.60 yuan per million tokens, and output at 7.50 yuan per million tokens.

• The page also lists higher prices for Pro; the configuration must select the corresponding price by model ID and must not charge ambiguously by the “Seed2.1” family name.

• The page's last update time is 2026-08-17 13:54:22.

Applicability boundaries

• The page does not provide complete request fields, tool-calling examples, reasoning tiers, response formats, or data-processing terms; unknown fields are explicitly marked as undisclosed/unverifiable.

• 256K is the input-length limit shown on the page; it does not guarantee that every gateway or Agent harness has the same usable context.

• Unit prices do not include routing, retries, preprocessing, human review, or failed-task costs; whether Turbo is cheaper must be calculated per completed task.

Source excerpts or observations (for compliant short quotations only)

• The official table separates online and Batch into different columns; the two execution modes must be kept separate in the internal ledger.

• The page notes that the actual purchasable specifications and fees are based on the order result.