# Bytedance Seed 1.8

Bytedance Seed 1.8 is a unified agentic model built to navigate search, code, and graphical user interfaces autonomously within one architecture.

- **Model ID:** `bytedance/seed-1.8`
- **Type:** chat
- **Providers:** bytedance
- **Context window:** 256,000
- **Maximum output tokens:** 64,000
- **Pricing:** $0.25/1M input tokens, $2/1M output tokens
- **Canonical page:** https://vercel.com/ai-gateway/models/seed-1.8

## Supported parameters

Detailed capability metadata has not been reported for this model.

## Example

```ts
import { streamText } from 'ai'

const result = streamText({
  model: 'bytedance/seed-1.8',
  prompt: 'Why is the sky blue?'
})
```

## About

Bytedance Seed 1.8 integrates three agent functions into one system. The Search Agent handles information retrieval across web and document sources. The Code Agent writes, debugs, and runs code. The GUI Agent interacts with graphical interfaces on desktop, web, and mobile using native vision rather than scripted automation, and operates software the way a human would.

Visual token efficiency is a core engineering focus. Bytedance Seed 1.8 reduces image encoding token requirements without sacrificing reasoning quality. This matters for GUI-heavy workloads where dozens of screenshots may pass through a single session. Three adaptive thinking modes calibrate processing depth to task complexity. They skip unnecessary compute on straightforward steps and use deeper reflection on ambiguous decisions.

In ByteDance's published benchmarks, Bytedance Seed 1.8 reaches 67.6 on BrowseComp-en, 87.8 on VideoMME (long-form video understanding), and 11.0 on ZeroBench (multimodal reasoning). It scores 62.0 on VLMsAreBiased and 47.2 on WorldTravel, up from its predecessor Seed 1.5-VL in those tables. Evaluations cover simulated workflows including travel planning, financial analysis, and software engineering. See https://docs.byteplus.com/en/docs/ModelArk/2123228 for methodology, tables, and comparisons.

## What to consider

For agentic pipelines with repeated GUI observation steps, confirm that your provider supports streaming for incremental processing of long action sequences. Compare token pricing ($0.25 in, $2 out per million tokens when listed).

## When to use

### Best For

- Browser and desktop automation where the GUI Agent observes, clicks, types, and navigates interfaces without custom scripting
- Research workflows that combine live information retrieval with synthesis and code execution in a single agent loop
- Agentic programming tasks where the model writes, tests, and iterates on code autonomously
- Long-form video understanding and analysis, supported by an 87.8 VideoMME score in ByteDance's published results
- Business process automation across financial analysis, itinerary generation, and document-heavy enterprise workflows

### Consider Alternatives When

- Your task is single-turn generation with no agentic component, and a lighter model would cost less
- You need deterministic tool-call schemas with strict JSON validation and must verify multi-step function calling before production
- The workflow is purely text-based with no visual inputs, and the model's image encoding overhead would waste compute
- You need a reasoning specialist optimized for mathematics or formal logic rather than generalized agent behavior

## Best for

- **Browser and desktop automation:** The GUI Agent observes, clicks, types, and navigates interfaces without custom scripting
- **Research workflows:** Combine live information retrieval with synthesis and code execution in a single agent loop
- **Agentic programming:** The model writes, tests, and iterates on code autonomously
- **Long-form video understanding:** Supported by an 87.8 VideoMME score in ByteDance's published results
- **Business process automation:** Financial analysis, itinerary generation, and document-heavy enterprise workflows

## Consider alternatives

- **Single-turn generation:** A lighter model would cost less when no agentic component is needed
- **Strict JSON validation:** Deterministic tool-call schemas require verification of multi-step function calling before production
- **Pure text workloads:** The model's image encoding overhead wastes compute when no visual inputs are involved
- **Formal reasoning specialists:** A model optimized for mathematics or formal logic may suit better than a generalized agent

## Frequently asked questions

### What does "generalized agentic model" mean for Bytedance Seed 1.8?

It completes multi-step tasks autonomously across search, code, and graphical interfaces. It covers all three instead of specializing in one capability.

### How does the GUI Agent in Bytedance Seed 1.8 work without traditional scripted automation?

It uses native vision to interpret screenshots and decide which actions to take: clicks, keystrokes, and form entries. It adapts to any UI layout without pre-defined selectors or automation scripts.

### What is the BrowseComp-en benchmark and why is Bytedance Seed 1.8's score notable?

BrowseComp-en tests retrieval and synthesis through web browsing in English. Bytedance Seed 1.8 scores 67.6 in ByteDance's published table. See https://docs.byteplus.com/en/docs/ModelArk/2123228 for the full benchmark context.

### How does token-efficient visual encoding benefit agentic applications?

Each GUI observation step sends one or more screenshots to the model. Fewer tokens per image means more steps fit within the context window at lower cost. This is especially important for long automation sessions with many intermediate observations.

### Does Bytedance Seed 1.8 support video understanding in addition to image inputs?

Yes. Bytedance Seed 1.8 scores 87.8 on VideoMME (long-form video understanding) in ByteDance's published results. It processes temporal sequences of visual content alongside text instructions.

### What kinds of real-world workflows was Bytedance Seed 1.8 evaluated on?

Evaluations cover simulated practical scenarios including travel planning, financial and business analysis, software engineering tasks, and multi-step information retrieval.

### Is Bytedance Seed 1.8 accessible without setting up a Volcano Engine account?

Yes. Through AI Gateway, you authenticate with an API key or OIDC token and route requests to Bytedance Seed 1.8. You don't need a separate Volcano Engine account.

## Links

- [Model page](https://vercel.com/ai-gateway/models/seed-1.8)
- [AI Gateway documentation](https://vercel.com/docs/ai-gateway)
- [Provider model documentation](https://docs.byteplus.com/en/docs/ModelArk/2123228)
- [Provider pricing](https://docs.byteplus.com/en/docs/ModelArk/1330310)


If you encounter incorrect information, can’t find what you need, need a capability Vercel doesn’t currently offer, or have any other feedback to report, please do it by following the instructions at https://vercel.com/feedback.md.