Seedream Historical Versions

This page lists all Seedream versions still callable on APIYI. All versions are active simultaneously with largely compatible parameters — in most cases you switch by changing only the model field (5.0 Flash and 5.0 Pro do not support sequential generation or streaming; see the compatibility matrix below). For the latest version and a general introduction, see Seedream Overview.

Versions at a Glance

Version ID	Release Date (UTC+8)	APIYI Price	Status	Recommended Use
`seedream-5-0-flash-260915`	2026-09-15	$0.018 / request	🆕 Latest · fast tier	High-volume generation, poster layouts, e-commerce and ad assets (~15–20 s per image, cheapest in the series)
`seedream-5-0-pro-260628`	2026-06-28	$0.12 / request	Pro tier	Top image quality / complex instructions for professional work (~2 min per image)
`seedream-5-0-260128`	2026-01-28	$0.035 / image	✅ Recommended	Best overall experience, text rendering, png output
`seedream-4-5-251128`	2025-11-28	$0.04 / image	✅ Recommended	4K + strongest text rendering (posters, ads)
`seedream-4-0-250828`	2025-08-28	$0.03 / image	🟡 Maintained	4K + best price, only version supporting prompt fast mode

seedream-5-0-260128 can also be invoked with the alias seedream-5-0-lite-260128 — the official docs accept both model_id strings, with identical behavior.

Per-version Specs

`seedream-5-0-flash-260915` (5.0 Flash)

Release date: 2026-09-15 (UTC+8); available on APIYI since 2026-09-24
APIYI price: $0.018 / request (fixed per-request price, 1 image per request), the same as BytePlus official pricing. Reference images are free (1, 3 and 10 reference images each billed $0.018 in our tests), and failed requests such as parameter errors are not billed. With top-up bonuses the effective price is roughly 9%–17% below the official rate
Resolution tiers: presets 1K / 1.5K / 2K (no 3K/4K; passing them returns 400). Omitting size defaults to 2K. With presets, the model picks the aspect ratio from the content (2K produced 1776×2368 and 1664×2496 portraits in our tests); pass exact pixels if you need a fixed ratio
Exact pixel range: 921,600–4,624,220 total pixels (about 0.92M–4.62M), long-to-short side ratio up to 16:1; out-of-range values return 400 with the valid range in the error message (2688x1712 and 4096x256 work)
Output formats: png / jpeg
Speed: 11–20 s for text-to-image; about 16 s for editing with reference images, about 28 s with 10 reference images
Core features:
- Positioned by BytePlus as fast, layout-strong and low-cost: better font variety, composition and color brightness than 5.0 Pro on layout tasks, with less gray or yellow color cast
- Accurate text rendering: a three-line Chinese poster and an English / Japanese / Korean title all rendered correctly in our tests
- Layer separation (transparent layers): layer_decomposition: true splits an image into a background base plus separate elements on alpha-channel layers, useful for re-layout and asset reuse
- Interactive editing: edit by position (for example, “add a NEW badge in the top-right corner”) while leaving the rest of the image unchanged
- Multi-image fusion with up to 10 reference images (an 11th returns 400)
Known limitations:
- No sequential_image_generation / stream: as with 5.0 Pro, any value (including "disabled") returns 400, so don’t send them
- n is silently ignored; you get and pay for 1 image
- Adds an “AI generated” watermark by default; pass watermark: false explicitly for commercial use
- Transparency needs layer separation, not prompting: asking for a transparent background in the prompt returns a png without an alpha channel (the checkerboard is drawn into the image). For transparent assets use layer_decomposition: true with one reference image; it returns a filled-in background base plus several RGBA layers (12 images in about 78 s at 1K in our tests). Layer separation is billed per output image: 12 images cost $0.216. A 2K layer request returned nothing after 15 minutes but was still billed $0.216, so always use 1K. See Image Editing
- For photorealistic portraits and fine material detail, BytePlus still recommends 5.0 Pro

`seedream-5-0-pro-260628` (5.0 Pro)

Release date: 2026-06-28 (UTC+8); available on APIYI since 2026-07-19
APIYI price: $0.12 / request (fixed per-request price, 1 image per request, ≈ ¥0.84 list). Officially the model uses two output-pixel price tiers (≤2.36M / >2.36M) plus a fee for reference images beyond the first; APIYI simplifies this to a flat per-request price with input-image fees included
Resolution tiers: presets 1K / 2K (no 3K/4K presets) plus exact pixels WxH up to 4.19M total (max 2048×2048; at 16:9 the longest edge reaches 2720x1530 ≈ 2.7K, verified)
Output format: png / jpeg
Prompt optimization: standard / fast
Highlights:
- Strongest image quality and instruction following in the family; interactive editing (specify edit locations via coordinates, selection boxes, arrows)
- Multi-image fusion officially supports up to 10 reference images
Known limitations:
- sequential_image_generation / stream are not accepted — any value (including "disabled") returns 400; omit both parameters entirely
- Slow generation: consistently ~2 minutes per image (110-130s measured) — set client timeout ≥ 240s
- 3.4× the unit price of 5.0-lite; use 5.0-lite for everyday work

`seedream-5-0-260128` (5.0-lite)

Release date: 2026-01-28 (UTC+8)
APIYI price: $0.035 / image (≈ ¥0.245 list)
Resolution tiers: 2K / 3K (no 4K)
Output format: png / jpeg (only version with png output)
Prompt optimization: standard
Highlights:
- Best overall experience; the most mature implementation of multi-image fusion / editing / batch sequence
- Only version supporting png output (and therefore transparent backgrounds)
- Mature streaming output (stream: true)
Known limits:
- Resolution capped at 3K (~3072×3072). For 4K assets, use 4.5 / 4.0
Official docs: docs.byteplus.com/en/docs/ModelArk/1824121

`seedream-4-5-251128`

Release date: 2025-11-28 (UTC+8)
APIYI price: $0.04 / image (≈ ¥0.28 list)
Resolution tiers: 2K / 4K
Output format: jpeg
Prompt optimization: standard
Highlights:
- 1.2B-parameter unified generation-editing architecture
- Text rendering breakthrough: small text remains crisp — best in class for posters, ads, UI screenshots
- Multi-image fusion explicitly supports up to 10 reference images
- Editing preserves lighting, tone, and natural facial detail
Known limits:
- jpeg output only (no png, no transparency)
News article: Seedream 4.5 launch

`seedream-4-0-250828`

Release date: 2025-08-28 (UTC+8)
APIYI price: $0.03 / image (≈ ¥0.21 list)
Resolution tiers: 1K / 2K / 4K (broadest coverage)
Output format: jpeg
Prompt optimization: standard / fast
Highlights:
- Battle-tested stable version
- Strong visual consistency, balanced 4K detail
- Only version supporting prompt fast mode — faster generation for cost-sensitive scenarios
Known limits:
- Text rendering weaker than 4.5
- jpeg output only

Migration Guidance

Assess differences

The versions are largely compatible at the parameter level — in most cases change only the model field to switch. Double-check:

Whether your size tier is supported by the new version (5.0 Flash and 5.0 Pro top out around 2K; 5.0 has no 1K / 4K; 4.5 has no 1K; 4.0 has all)
Whether you send sequential_image_generation or stream (remove them when switching to 5.0 Flash / 5.0 Pro, or you get a 400)
Whether you depend on output_format: "png" (5.0 only)
Whether you use prompt_optimization: "fast" (4.0 only)

Run them side by side

Run the same prompt batch on the old and new versions, compare quality and cost. Validate with a small batch (10-20 images) before scaling.

Roll out gradually

Cut traffic in stages (10% / 50% / 100%). At each stage, observe quality, failure rate, and cost before scaling further.

Keep a fallback

Keep both old and new model configs in production code. If the new version misbehaves, flip back instantly. Each version bills at its own unit price — running both has no extra cost.

Legacy Invocation Example

{/* Switching versions only changes the model field; other params are compatible */}
from openai import OpenAI

client = OpenAI(api_key="sk-your-api-key", base_url="https://api.apiyi.com/v1")

resp = client.images.generate(
    model="seedream-4-0-250828",   # change this line to switch to 4.5 / 5.0
    prompt="A serene mountain landscape at golden hour, snow-capped peaks, ultra detailed, 4K",
    size="4K",                      # NOTE: 5.0 does not support 4K — use 2K or 3K
    response_format="url",
    extra_body={
        "watermark": False,
    }
)

print(resp.data[0].url)

Cost Comparison

Estimated by typical volume (before top-up bonuses; effective price drops to as low as 80% with bonuses):

Version	Unit Price	100 images	1,000 images	10,000 images
`seedream-5-0-flash-260915`	$0.018 / request	$1.8	$18	$180
`seedream-5-0-pro-260628`	$0.12 / request	$12	$120	$1200
`seedream-5-0-260128`	$0.035	$3.5	$35	$350
`seedream-4-5-251128`	$0.04	$4	$40	$400
`seedream-4-0-250828`	$0.03	$3	$30	$300

How to choose:

4K assets + strong text → 4.5
4K assets + best price → 4.0
High volume / poster layouts / speed and cost → 5.0 Flash ($0.018/request, ~15–20 s per image, up to about 2K)
Best overall / png output / transparent backgrounds → 5.0
Long-running stable batch production → 4.0 (proven, cheapest, fast prompt mode)
Top quality / complex instructions for professional work → 5.0-pro ($0.12/request + ~2 min per image — skip unless you need it)

Compatibility Matrix

Dimension	5.0-flash	5.0-pro	5.0	4.5	4.0	Migration note
`1K` resolution tier	✅	✅	❌	❌	✅	Migrating from 4.0 to 5.0-lite → switch 1K to 2K
`4K` resolution tier	❌	❌	❌	✅	✅	Migrating from 4.5/4.0 to the 5.0 series → change the tier
`output_format: "png"`	✅ (no alpha when generating directly; transparency via layers)	✅	✅	❌	❌	Migrating from the 5.0 series → 4.5/4.0: transparency is lost
`prompt_optimization: "fast"`	❌ (not supported officially)	✅	❌	❌	✅	5.0-lite / 4.5 don’t support fast — drop the parameter when switching
`image` array (multi-image fusion)	✅ (up to 10, tested)	✅ (up to 10 explicit)	✅	✅ (up to 10 explicit)	✅	Identical protocol
`sequential_image_generation`	❌ (any value returns 400, tested)	❌ (any value returns 400)	✅	✅	✅	Migrating to flash / pro: remove the parameter entirely (including `"disabled"`)
`stream` streaming	❌ (400 if passed, tested)	❌ (400 if passed)	✅	✅	✅	Migrating to flash / pro: remove the parameter
`layer_decomposition` layers	✅ (billed per output image, use 1K)	—	—	—	—	Tested on 5.0-flash only
Response fields (`url` / `b64_json`)	identical	identical	identical	identical	identical	—
Billing	$0.018 per request	$0.12 per request	per image	per image	per image	—

Seedream Overview
Text-to-Image Playground
Image Editing Playground
Seedream 4.5 launch
BytePlus official tutorial: docs.byteplus.com/en/docs/ModelArk/1824121

Basics

Basic API

Image API (Official)

Video API (Official)

Multimodal Understanding API

Text Generation

Audio API (Official)

Retrieval & Safety

Versions at a Glance

Per-version Specs

`seedream-5-0-flash-260915` (5.0 Flash)

`seedream-5-0-pro-260628` (5.0 Pro)

`seedream-5-0-260128` (5.0-lite)

`seedream-4-5-251128`

`seedream-4-0-250828`

Migration Guidance

Legacy Invocation Example

Cost Comparison

Compatibility Matrix

​Versions at a Glance

​Per-version Specs

​seedream-5-0-flash-260915 (5.0 Flash)

​seedream-5-0-pro-260628 (5.0 Pro)

​seedream-5-0-260128 (5.0-lite)

​seedream-4-5-251128

​seedream-4-0-250828

​Migration Guidance

​Legacy Invocation Example

​Cost Comparison

​Compatibility Matrix

​Related Docs

Versions at a Glance

Per-version Specs

`seedream-5-0-flash-260915` (5.0 Flash)

`seedream-5-0-pro-260628` (5.0 Pro)

`seedream-5-0-260128` (5.0-lite)

`seedream-4-5-251128`

`seedream-4-0-250828`

Migration Guidance

Legacy Invocation Example

Cost Comparison

Compatibility Matrix

Related Docs