← Back to hard AIs

Verify critical details — pricing, licensing, availability — with the model's source before business decisions. Full methodology →

Models

ByteDance

4.0 ★ — On the strength of the open Seed-OSS line — a genuinely strong Apache 2.0 model with a 512K context window. This is an initial entry focused on Seed-OSS; the broader ByteDance lineup (Doubao and others) is not yet cataloged, so the rating reflects only what's covered so far.

Type
big-tech-lab
Country
CN
Founded
2012
License posture
mixed
Website
https://seed.bytedance.com

Quick Take

The Chinese tech giant behind TikTok, increasingly active in open AI through its Seed research team — most notably the Apache 2.0 Seed-OSS models with their 512K-tokenThe basic unit of text a model reads and writes. Tokens are roughly three-quarters of a word in English — so 100 tokens is about 75 words. Models don't see letters or words directly; they see tokens. Pricing is almost always quoted per million tokens, and context windows are measured in tokens rather than words. context windows.

Who They Are

ByteDance is the China-based technology company best known globally for TikTok. Its AI research runs through the ByteDance Seed team, formed in 2023, which builds foundation models for both ByteDance's own products and the open-sourceA stricter standard than open-weight: the weights, the training code, and the training data are all released publicly. Very few large language models meet the full open-source bar — most "open" models in the AI world are actually open-weight. When in doubt, check the license file and the creator's documentation. community. The team's open releases ship on Hugging Face and GitHub, and its most prominent open model to date is Seed-OSS-36B — a 36-billion-parameter model with a notably large 512K-tokenThe basic unit of text a model reads and writes. Tokens are roughly three-quarters of a word in English — so 100 tokens is about 75 words. Models don't see letters or words directly; they see tokens. Pricing is almost always quoted per million tokens, and context windows are measured in tokens rather than words. context windowThe maximum amount of text the model can "see" at once — prompt plus prior conversation plus any documents you give it. Measured in tokens (which are roughly three-quarters of a word each). A 128K context window is about 96,000 words of input — roughly a 400-page book. Larger context windows let the model work with bigger documents but cost more to run., released in August 2025 under Apache 2.0.

For business readers, the relevant distinction is between ByteDance's open and closed work. The open Seed-OSS line is freely downloadable and self-hostable under a permissive license; ByteDance also runs closed, hosted products (such as the Doubao assistant, used heavily in China), which aren't cataloged here yet. This entry is an initial one focused on Seed-OSS as a foundation and fine-tuning base.

Model Philosophy

The open Seed-OSS releases follow the now-common Chinese-lab pattern of shipping capable models under permissive Apache 2.0 terms — competing on capability-per-parameter and accessibility rather than locking capability behind an API. Seed-OSS emphasizes long-context handling, reasoning, and agentic/tool-use capabilities, with a distinctive "controllable thinking budget" that lets users tune how much the model reasons before answering. The team has also signaled interest in research-friendly releases, publishing base models both with and without synthetic training data.

What To Know Before You Commit

The open Seed-OSS weightsThe numerical values inside a trained model that encode everything it has learned. A model is, functionally, a giant list of weights — tens of billions of numbers for a mid-sized model, hundreds of billions for a frontier model. "Open-weight" means those numbers are published. "Downloading the weights" means getting the actual file you'd need to run the model yourself. are Apache 2.0 — unrestricted commercial use, modification, and redistribution — and because you self-host them, they carry no data-routing concerns. That makes them straightforward to adopt despite ByteDance's China base. The main caveats are ordinary ones for this catalog's purposes: this is an initial ByteDance entry covering Seed-OSS specifically, so treat the broader lineup as not-yet-cataloged, and (as with any model) verify the current license on the specific checkpointA specific saved version of a model at a particular point in training. When a creator releases "Llama 3.1 8B Instruct," they're releasing a checkpoint — a frozen snapshot of the model as it existed at the end of training. Most models ship only a single public checkpoint; some creators release multiple (base, instruct, reasoning variants of the same underlying model). you download.

Original Models

Seed 2 0

Seed 2.0's coding specialist — code generation, debugging, and PR-style review, integrated with ByteDance's TRAE editor; ByteDance claims near-Claude coding at a fraction of the price. API-only on Volcano Engine.

The balanced middle tier of ByteDance's Seed 2.0 line — most of Pro's multimodalA model that can handle more than one type of input or output — typically text plus images, sometimes plus audio or video. "GPT-4 Vision" and "Llama 3.2 11B Vision" are multimodal models that accept both text and images. A text-only model is called "unimodal" but nobody uses that term; text-only is the assumed default. ability at lower cost, API-only on Volcano Engine, aimed at everyday production work.

The fastest, cheapest Seed 2.0 tier — built for high-throughput, latency-sensitive batch tasks where Pro-level reasoning isn't needed. API-only on Volcano Engine.

ByteDance's frontier model and the engine behind the Doubao assistant — a multimodalA model that can handle more than one type of input or output — typically text plus images, sometimes plus audio or video. "GPT-4 Vision" and "Llama 3.2 11B Vision" are multimodal models that accept both text and images. A text-only model is called "unimodal" but nobody uses that term; text-only is the assumed default. reasoner with a 256K context that benchmarks near GPT-5 and Gemini-3 class while costing a fraction as much, with t

Seedance

ByteDance's proprietary text-to-video model, known for highly realistic clips — and for drawing cease-and-desist letters from Disney and Paramount in Feb 2026 over alleged training on their works. Capable, but carries unresolved IP risk.

Seedream

ByteDance's proprietary text-to-image model — high-detail output up to 2K native (4K enhanced) with strong long-text rendering, delivered through CapCut and Volcano Engine rather than as downloadable weightsThe numerical values inside a trained model that encode everything it has learned. A model is, functionally, a giant list of weights — tens of billions of numbers for a mid-sized model, hundreds of billions for a frontier model. "Open-weight" means those numbers are published. "Downloading the weights" means getting the actual file you'd need to run the model yourself..

Seed Oss

ByteDance's open-weightA model where the trained weights are freely downloadable — you can run it yourself without contacting the creator. Llama, Mistral, Qwen, and Gemma are open-weight. Open-weight does not mean open-source: the training data and code often stay private. The license still governs what you can do with the weights, including whether you can use them commercially. flagship-by-proxy: a 36-billion-parameter Apache-2.0 model with a native 512K context — four times the open-sourceA stricter standard than open-weight: the weights, the training code, and the training data are all released publicly. Very few large language models meet the full open-source bar — most "open" models in the AI world are actually open-weight. When in doubt, check the license file and the creator's documentation. norm — built for long-document reasoning and agent workflows

Seed Coder

A small, open-weightA model where the trained weights are freely downloadable — you can run it yourself without contacting the creator. Llama, Mistral, Qwen, and Gemma are open-weight. Open-weight does not mean open-source: the training data and code often stay private. The license still governs what you can do with the weights, including whether you can use them commercially. code model from ByteDance's Seed team, with a reasoning-tuned variant — a self-hostable option for code generation and debugging on consumer hardware.

Seed 1

ByteDance's prior-generation Seed line, superseded by Seed 2.0 but still sold as a budget API tier — the Seed 1.6 Flash variant is one of the cheapest models on Volcano Engine (around $0.02 per 1M input tokens).

Bagel

An open-weightA model where the trained weights are freely downloadable — you can run it yourself without contacting the creator. Llama, Mistral, Qwen, and Gemma are open-weight. Open-weight does not mean open-source: the training data and code often stay private. The license still governs what you can do with the weights, including whether you can use them commercially. "unified" multimodalA model that can handle more than one type of input or output — typically text plus images, sometimes plus audio or video. "GPT-4 Vision" and "Llama 3.2 11B Vision" are multimodal models that accept both text and images. A text-only model is called "unimodal" but nobody uses that term; text-only is the assumed default. model — one model that both understands and generates text, images, and video — under a permissive Apache 2.0 license. Useful as a single self-hostable building block for mixed media tasks.

Ui Tars

An open-weightA model where the trained weights are freely downloadable — you can run it yourself without contacting the creator. Llama, Mistral, Qwen, and Gemma are open-weight. Open-weight does not mean open-source: the training data and code often stay private. The license still governs what you can do with the weights, including whether you can use them commercially. "GUI agent" model that sees a screen and operates software like a person — clicking, typing, navigating — for computer-use automation. 7B, runnable locally, with a larger 72B variant available.

Sources