讯飞星火
科大讯飞A personalized AI assistant built on a large model trained entirely on domestic computing power, supporting Q&A, content creation, image generation, and code generation
A general-purpose large model API with 1M context and 256k output, handling text, image, and video inputs with built-in web search.
Doubao-Seed-2.1-pro/260915 is a high-performance general-purpose large model launched by Volcano Ark, offering a 1M context window and 256k maximum output. Input supports text, images, and video (audio not supported), with native Web Search built in. It is integrated via the Volcano Ark platform in the form of API, supporting three interfaces—Chat / Response / Messages—along with structured output, implicit and explicit caching, batch inference, and four reasoning depth tiers from minimal to high. It is well suited for enterprise developers who need long-document analysis, long-report generation, structured data extraction, or low-latency production-grade inference services. Pricing is billed per million tokens, and the API Key must be created in the console and stored securely.
Core technical specifications (source: Volcano Engine article): context window 1M, maximum input 1M, maximum output 256k, TPOT 15-18ms, toB default rate limits of RPM 500 and TPM 1000k. Input types are text, images, and video; audio is explicitly not supported (including audio within video); output type is text. Reasoning mode can be set to minimal/low/medium/high (default: high), and image understanding detail is divided into low/high/xhigh (default: high). The API address is https://ark.cn-beijing.volces.com/api/v3, with authentication via Bearer Token in the HTTP header. The Response API is the most feature-rich, natively supporting Web Search, structured output, caching, and tool calling; the Messages API is suited for multi-turn conversations and tool calling; the Chat API is suited for standard conversations. The cache-hit price is 1.2 yuan per million tokens, roughly one-fifth of the standard input price of 6.0 yuan per million tokens. The documentation also notes integration with third-party coding tools such as Claude Code, OpenClaw, and Codex CLI.
Enterprise developers who need to process ultra-long documents (contract analysis, financial report interpretation, codebase comprehension), generate ultra-long outputs, build low-latency production-grade API services, or create structured data extraction or web-search-enabled intelligent applications; can also configure this model endpoint in coding tools.
Register a Volcano Engine account and complete real-name verification.
Go to the Volcano Ark console and enable the Doubao-Seed-2.1-pro model service.
Create an inference endpoint and obtain the Endpoint ID.
Create a new API Key in the console's "API Key Management" section, configure permissions and IP allowlists as needed, then copy and save it (displayed only once).
Choose one of the three API types—Chat / Response / Messages—and send requests to https://ark.cn-beijing.volces.com/api/v3, authenticating with Authorization: Bearer <API_KEY>.
Configure reasoning_effort depth, response_format structured output, or the web_search tool as needed.
For long-document scenarios, we recommend summarizing and extracting first before deeper analysis, and using prefix cache to cache document prefixes and reduce repeated inference costs.
Requires registering a Volcano Engine account and completing real-name verification (individual or enterprise); the API Key is displayed only once and must be stored securely; toB default rate limits are RPM 500 and TPM 1000k; audio input is not supported (including audio within video); PDFs of hundreds of pages must be converted to text first; the source does not specify regional or device restrictions—check the official website for details.
费用与套餐:Uniform pricing within the 1M window: input 6.0 yuan/million tokens, output 30.0 yuan/million tokens, cache hit 1.2 yuan/million tokens. The source notes that "latest pricing is subject to the official Volcano Engine pricing page"; prices may change, so verify the official pricing page before purchase.