HomeAI models
ByteDance Seed: Seed-2.0-Lite
Bytedance
•
Seed-2.0-Lite
•
Released 
February 14, 2026

Bytedance
Seed-2.0-Lite

ByteDance's mid-tier Seed 2.0 model for image and video understanding, tool use and reasoning in high-volume production.

Modality:
Text
Image
PDF
model ID
bytedance-seed/seed-2.0-lite
Start for free
arrow black icon
View API Docs
arrow white icon

Output Speed *

222.15
tok/s

Intelligence Index *

8.7
/ 100

Context Window *

262144
tokens

Input price

1.5
Anytoken

Output price

12
Anytoken
Seed-2.0-Lite: Video-Capable Reasoning and Agents at Production Scale Seed-2.0-Lite is ByteDance Seed's mid-tier model in the Seed 2.0 series, sitting between Pro and Mini as the general-purpose production option. It accepts text, image and video input, returns text, and offers configurable reasoning, function calling and a 262,144-token context window. ByteDance's late-April upgrade added audio understanding. It suits multimodal extraction, video analysis and tool-using agents where response speed and cost matter more than top-tier coding scores. Access Seed-2.0-Lite through AnyAPI.ai

Performance

Agentic Coding Scores That Hold Up for a Mid-Tier Model

In ByteDance's Seed2.0 model card, Seed2.0 Lite scores 73.5 on SWE-Bench Verified and 45.0 on Terminal Bench 2.0. That is ahead of GPT-5-mini High on both and behind Gemini 3 Flash High. These are vendor-reported results, not independent measurements. They suggest Lite can handle repository-level bug fixing and command-line tasks without leading its efficient-model class. In production it fits agent loops that mix tool calls with image or video inputs, with a stronger coding model held back for the hardest tasks.

Technical Specifications

What the model supports

Seed-2.0-Lite takes text, image and video input and returns text only. Its context window is 262,144 tokens, and OpenRouter lists a maximum output of 131,072 tokens. Some gateways list lower output caps, so confirm the limit on your endpoint. Reasoning can be tuned or disabled through provider parameters. The large output ceiling matters for long structured extraction and report generation. The 262K input window lets long documents fit in a single request.
Verified Specifications — 
Seed-2.0-Lite
*
Input modalities
Text
Image
PDF
output modalities
Text
Context window
262144
 tokens
Maximum output tokens
131072
Reasoning
Yes
Knowledge cutoff
February 14, 2026
Pricing (standard)
1.5
 AnyTokens in
 / 
12
 AnyTokens out

Limitations & Trade-offs

Where Seed-2.0-Lite falls short

1
Behind Gemini 3 Flash on agentic coding. In ByteDance's own model card, Seed2.0 Lite scores 45.0 on Terminal Bench 2.0 and 41.1 on Multi-SWE-Bench. Gemini 3 Flash High scores 60.0 and 59.0. Terminal-driven coding agents and multi-repository refactoring are therefore weaker fits. For those workloads, test a stronger coding model, or the dedicated Seed 2.0 Code model, before defaulting to Lite.
2
Text-only output. Seed-2.0-Lite accepts text, image and video, and ByteDance's April update added audio understanding. It returns text only. It cannot generate images, audio or video. Audio input has not been confirmed on every endpoint, so pipelines that depend on audio should verify support on the specific gateway before committing.
3
Benchmarks are vendor-reported. The sources reviewed contained no independent Seed 2.0 Lite scores; the Artificial Analysis ByteDance Seed page lists only two benchmarked models, neither of them this one. ByteDance has also since released a Seed 2.1 generation. Teams that need independently verified quality should build their own evaluation set rather than rely on the model card.
4
Inconsistent output caps across endpoints. OpenRouter lists a 131,072-token maximum output, but other gateways list 128,000, 65,536 or as low as 8,192 for the same model name. This matters for long reports, large JSON extractions and code generation. Check the limit and default max_tokens on your endpoint before relying on very long single responses.

Best-Fit Workloads

Where this model earns its place

01

Video and image understanding

‍
Seed-2.0-Lite accepts image and video input, and ByteDance positions the Seed 2.0 series around multimodal perception. For its latest version, ByteDance reports state-of-the-art results on several video and audio understanding benchmarks. These are vendor claims. Content tagging, scene classification, media review queues and visual question answering benefit most. Output is text only, and audio input should be confirmed on your endpoint.

02

Tool-using agents

‍
The model supports function calling and configurable reasoning, and ByteDance describes the Seed2.0 series as built for long-horizon agent tasks. Its 73.5 on SWE-Bench Verified (vendor-reported) shows usable engineering ability. It suits customer-intake flows, workflow automation and search agents. For terminal-heavy coding agents, a stronger coding model is likely better.

03

Structured extraction from unstructured documents

‍
A 262,144-token window, image input and JSON Schema output support make Seed-2.0-Lite suitable for pulling fields from long contracts, scanned forms and mixed-media files. Provider positioning lists unstructured information processing as a core use, and ByteDance reports strong long-document results for the Seed2.0 family. Validate schema adherence on your own documents, since structured output behavior varies by gateway.

04

Search, recommendation and data analysis pipelines

‍
Provider descriptions position Seed-2.0-Lite for high-frequency enterprise tasks such as content creation, search and recommendation, and data analysis. The lower-cost tier and adjustable reasoning effort let teams use light reasoning on routine requests and raise effort only for harder queries. It is a better default for volume than for frontier-difficulty reasoning.

Pricing in anytokens via AnyAPI
Input
1.5
₳
Output
12
₳
Cache write
—
₳
Cache read
—
₳
Get API access
arrow black icon
Read Docs

on this page

PerformanceTechnical SpecificationsLimitations & Trade-offsBest-Fit WorkloadsIntegrationFAQ

Integration

Access Seed-2.0-Lite via AnyAPI.ai

Access Seed-2.0-Lite through AnyAPI.ai using a unified API built for multi-model AI applications. Integrate Seed-2.0-Lite without maintaining a separate provider-specific connection, and keep the flexibility to test, switch, or combine models as your application requirements evolve.

01

One API integration

Access Seed-2.0-Lite and other AI models through the same API workflow instead of maintaining separate integrations for every provider.

02

Easy model switching

Test Seed-2.0-Lite against alternative models or switch models as your performance, capability, or cost requirements change without rebuilding your application around another provider API.

03

Flexible for production

Use Seed-2.0-Lite from experimentation through production while keeping your AI stack flexible as workloads, traffic, and model requirements evolve.

04

Multi-model applications

Use Seed-2.0-Lite for the workloads where it performs best and combine it with other models for tasks that require different capabilities, performance, or efficiency.

Start free
arrow black icon
Read Docs

Frequently Asked Questions

Answers to common questions about integrating and using this AI model via AnyAPI.ai

Seed-2.0-Lite has a context window of 262,144 tokens, which ByteDance's BytePlus channel describes as 256K. OpenRouter lists a maximum output of 131,072 tokens. Output limits differ by gateway, with some endpoints listing lower caps, so check the limits on the endpoint you use.

Yes. Seed-2.0-Lite accepts text, image and video input and returns text. ByteDance's late-April 2026 upgrade added audio understanding. It does not generate images, audio or video. Confirm audio support on your specific API endpoint before building audio-dependent workflows.

Yes. Seed-2.0-Lite supports a thinking mode that is enabled by default on at least one provider endpoint. That endpoint also exposes reasoning_effort with low, medium and high levels, and thinking can be turned off. Parameter names and defaults may vary by gateway, so check your provider's documentation.

Seed-2.0-Lite is the mid-tier model in the Seed 2.0 series. ByteDance describes it as balancing output quality and response speed for general-purpose production use. Pro is the larger flagship variant, and Mini is the lightest option, aimed at low-latency, high-concurrency, cost-sensitive use. Choose Lite as the default and move up or down by workload.

Seed-2.0-Lite is a capable but not leading coding model. ByteDance's model card reports 73.5 on SWE-Bench Verified and 45.0 on Terminal Bench 2.0, below Gemini 3 Flash High on both. These are vendor-reported figures. For dedicated agentic coding, consider Seed 2.0 Code or another coding-focused model.

* Benchmark data source: Artificial Analysis artificialanalysis.ai