Seed 2.0 Lite: Puter.js/OpenAI-Compatible Calls and Multimodal Configuration

One-sentence takeaway

Within the Puter ecosystem, you can connect using the model slug bytedance-seed/seed-2.0-lite through Puter.js or the OpenAI SDK. This configuration is suitable for quickly validating text, image, video, and tool calls, but its pricing, context window, and publication date are provider-specific and should not be treated as official Ark specifications.

Use cases

• Suitable tasks: Rapid prototyping, frontend/Node integration, document and image processing, Agent tool calls, and Puter.js applications that do not want to manage a separate API key first.

• Unsuitable tasks: Directly migrating Puter's free/user-paid model, USD pricing, or 262K context window to Volcengine Ark or another provider.

• Applicable model version: Puter slug bytedance-seed/seed-2.0-lite; the page does not disclose the exact upstream snapshot.

• Applicable clients, Agents, or APIs: Puter.js, Puter OpenAI-compatible API, Python OpenAI SDK, curl.

• Recommended reasoning tier and parameters: The page does not disclose a fixed reasoning tier; record reasoning, max_tokens, temperature, and tool configuration on the actual provider.

Ready-to-use content

Minimal Puter.js configuration (the code comes from the original page's call example; the API token and billing are determined by Puter's account policies):

// npm install @heyputer/puter.js
import { puter } from '@heyputer/puter.js';

puter.ai.chat(
  "Read the provided document or image, list the key fields first, then output them as JSON; return null when uncertain.",
  { model: "bytedance-seed/seed-2.0-lite" }
).then((response) => {
  console.log(response.message.content);
});

OpenAI-compatible configuration fields:

from openai import OpenAI

client = OpenAI(
    base_url="https://api.puter.com/puterai/openai/v1/",
    api_key="YOUR_PUTER_AUTH_TOKEN",
)

response = client.chat.completions.create(
    model="bytedance-seed/seed-2.0-lite",
    messages=[{"role": "user", "content": "Your task prompt"}],
)

Test/workflow steps

1. Start with the text request provided by the page to verify the slug, authentication, and response fields.

2. Then add an image, video, or tool schema according to the provider's supported method; the page explicitly promotes these capabilities but does not provide a complete multimodal request example.

3. For production tasks, save the provider, model slug, input modality, reasoning, max_tokens, tools, token usage, latency, and error logs.

4. Record Puter results separately from results using the same prompt on Ark or another provider, to avoid attributing gateway differences to the model.

Original evidence and data

• The page provides three calling methods—Puter.js, Python OpenAI SDK, and curl—and all use the model slug bytedance-seed/seed-2.0-lite.

• The page's model card shows a 262K context window, 131K maximum output, input pricing of $0.25 per million tokens, and output pricing of $2 per million tokens; these are the values shown on the Puter page.

• The page says Seed 2.0 Lite supports text, image, and video understanding and tool calls, and positions it as a production-grade default model; complete upstream API details are not public.

Scope and limitations

• The Puter page is a third-party integration/product page, not a ByteDance official model card. Its publication date (2026-03-10) differs from the official launch blog (2026-02-14), so the official source should take precedence.

• The page uses both “256K” and “262K” as approximate descriptions; context windows and maximum outputs from different providers cannot be directly interchanged.

• Do not treat the generic quantum-computing question in the example as a model evaluation; this entry provides call configuration, not a performance conclusion.

Source excerpts or observations (for compliant short quotations only)

• The page explicitly uses bytedance-seed/seed-2.0-lite as the sole calling slug.

• The page recommends connecting through Puter.js or an OpenAI-compatible API, which is suitable for low-cost initial integration validation.