Seedream 5.0 vs GPT Image 1.5 vs FLUX.2 [max]: Features, Pricing and Which Image Model Fits Which Job
A fact-checked comparison of ByteDance Seedream 5.0, OpenAI GPT Image 1.5 and Black Forest Labs FLUX.2 [max]: architecture and access, reference images and search grounding, editing, official pricing, licensing and availability, plus what has changed since early 2026.
Within about two months, from mid-December 2025 to mid-February 2026, three companies shipped image models that set the pace for commercial image generation: OpenAI’s GPT Image 1.5, Black Forest Labs’ FLUX.2 [max] and ByteDance’s Seedream 5.0. Each takes a different approach to access, editing and pricing. This guide compares them using the companies’ documentation and published prices rather than our own side-by-side tests, because results depend heavily on the prompts chosen. It also notes what has changed since, because all three lines have moved on.
At a glance
| Seedream 5.0 | GPT Image 1.5 | FLUX.2 [max] | |
|---|---|---|---|
| Developer | ByteDance (Seed team) | OpenAI | Black Forest Labs |
| Released | February 10, 2026 (Lite: February 13) | December 16, 2025 (snapshot gpt-image-1.5-2025-12-16) | December 2025 |
| Where to use it | CapCut, Jianying, Dreamina; API via BytePlus ModelArk | ChatGPT; OpenAI Images API | BFL Playground and API; partner platforms |
| Resolution | 2K native, 4K with AI enhancement | 1024x1024, 1024x1536, 1536x1024 | Up to 4 megapixels |
| Reference images | Multiple (varies by version) | Image input and edits | Up to 8 via API, 10 in Playground |
| Web search grounding | Yes (toggleable in 5.0 Lite) | Not part of the image model | Yes |
| Editing | Brush-based element editing; image-to-image | Edits and inpainting endpoint | Editing with multiple references |
| Open weights | No | No | No ([dev] and [klein] siblings are open) |
| Status in November 2026 | Succeeded by 5.0 Pro and Flash | Deprecated by OpenAI | Still Black Forest Labs’ top image model |
Sources: 36Kr on Seedream 5.0’s launch, ByteDance Seed’s Seedream 5.0 Lite announcement, OpenAI’s GPT Image 1.5 model page and Black Forest Labs’ FLUX.2 overview.
Seedream 5.0: reasoning, search and editing in creator apps
ByteDance launched Seedream 5.0 on February 10, 2026, first inside its own creative apps: Jianying and its overseas version CapCut, and the Xiaoyunque creation platform, with a limited free test on Jimeng (Dreamina). According to the launch coverage, it outputs 2K images directly and 4K with AI enhancement, supports retrieval-based generation for the first time, improves prompt understanding and texture detail, and adds a brush tool for selecting and adjusting specific elements.
Three days later the Seed team released Seedream 5.0 Lite, which it describes as a step towards a unified multimodal model. Its emphasis is not speed or resolution but understanding: interpreting vague instructions, capturing features of reference images, applying world knowledge to infographics, and a real-time search option that can be switched on for time-sensitive content or off for more stable results. ByteDance says its Elo score in internal evaluations exceeds Seedream 4.5.
Strengths. Text-rich images such as infographics and posters, Chinese and English text, reference-guided editing and style transfer, and tight integration with CapCut for creators who already edit video there.
Considerations. API access is through ByteDance’s BytePlus cloud outside China, whose terms limit availability in some regions including the United States, according to a pricing and access summary by GenMagic. Business users should check data-handling terms.
GPT Image 1.5: instruction following inside ChatGPT
OpenAI describes GPT Image 1.5 as having better instruction following and prompt adherence than its predecessors. Its biggest advantage was distribution: it powered image generation inside ChatGPT, so users could create and revise images conversationally (“make the background warmer”, “keep everything but change the shirt to blue”) without any technical setup. Developers accessed it through the Images API’s generation and edit endpoints, including inpainting.
Pricing. OpenAI prices it by quality and size. A 1024x1024 image costs $0.009 at low quality, $0.034 at medium and $0.133 at high; larger 1024x1536 images cost $0.013, $0.05 and $0.20 respectively, according to its model page. The spread matters: the same request can vary roughly fifteen-fold in price depending on the quality setting.
Status. OpenAI now lists GPT Image 1.5 as deprecated and has moved ChatGPT to newer image models. We compare the current options in Nano Banana, ChatGPT Images 2.5 or Adobe Firefly. Existing integrations should plan a migration.
FLUX.2 [max]: highest quality and reference consistency
FLUX.2 [max] is the top tier of Black Forest Labs’ FLUX.2 family, which also includes [pro], [flex], the open-weight [dev] and the small [klein]. The company positions [max] for final assets: the highest image quality and editing consistency in the family, up to 8 reference images through the API, output up to 4 megapixels and grounding search, which lets the model look up current information on the web when a prompt needs it.
Earlier claims that FLUX.2 [max] has only limited editing are wrong: every FLUX.2 variant combines text-to-image generation and multi-reference editing in one model, according to Black Forest Labs’ FLUX.2 announcement.
Pricing. The first output megapixel costs $0.07 and each additional megapixel $0.03, with reference images charged at $0.03 per megapixel, as listed on OpenRouter. We cover prompting and workflow in our FLUX.2 [max] guide.
Pricing compared
| Example | Seedream | GPT Image 1.5 | FLUX.2 [max] |
|---|---|---|---|
| Standard ~1MP image | BytePlus advertises pricing from $0.02 per image; per-model prices differ | $0.009 to $0.133 by quality | About $0.07 |
| Larger image | 5.0 Pro: $0.045 up to 1.5K, $0.09 above (per GenMagic’s reading of BytePlus prices) | Up to $0.20 at high quality | About $0.10 for 2MP, more with references |
The BytePlus Seedream page advertises per-image pricing starting at $0.02. Prices change frequently and differ on resellers, so treat these as orientation rather than a quote.
Which to choose
| If you need | Lean towards | Why |
|---|---|---|
| Text-heavy graphics, infographics, Chinese text | Seedream 5.0 line | Emphasis on text rendering and reasoning about layout |
| Creators already working in CapCut | Seedream 5.0 | Built into the editing app |
| Conversational revisions with no setup | OpenAI’s ChatGPT image tools (now newer models) | Iterate in plain language |
| Cheap drafts at volume | GPT Image 1.5 low quality (while available) or Seedream’s cheaper tiers | Low per-image prices |
| Photorealistic final assets with consistent products or characters | FLUX.2 [max] | Reference consistency and up to 4MP output |
| Visuals of current events or products | FLUX.2 [max] or Seedream with search | Both offer web grounding |
Before committing, run the same set of your own prompts through each model, including your hardest cases: brand text, specific products and consistent characters. Compare the results with the people who will use the images.
What has changed since early 2026
The landscape has moved quickly. ByteDance added Seedream 5.0 Pro in July 2026, with spatial annotation editing and layer separation, and a cheaper Flash variant. OpenAI deprecated GPT Image 1.5 in favour of newer ChatGPT image models. Black Forest Labs has kept FLUX.2 [max] as its top image model while expanding into video. If you are choosing today, compare these current versions rather than the originals.
Sources and further reading
- OpenAI: GPT Image 1.5 model page
- OpenAI: image generation guide
- Black Forest Labs: FLUX.2 overview
- ByteDance Seed: Seedream 5.0 Lite
- ByteDance Seed: Seedream 5.0 Pro
Our analysis
These three models illustrate three business models: ByteDance putting image generation inside its creator apps and cloud, OpenAI putting it inside a chat assistant, and Black Forest Labs selling a premium API alongside open-weight siblings. None was best at everything, and published claims rarely transfer directly to a specific brand’s needs. The sensible approach is to define your use cases, test current versions on your own prompts, and weigh price per usable image, licensing and regional availability alongside quality.