DALL-E 3 vs FLUX: Full Comparison and 2026 Status
DALL-E 3 no longer exists as an active product: OpenAI retired it on May 12, 2026 and replaced it with GPT Image 2, while FLUX remains an actively developed, open-weight model family from Black Forest Labs now in its FLUX.2 generation. Anyone comparing the two today is really choosing between FLUX and DALL-E 3’s actual successor, not the DALL-E 3 that existed through 2025. This guide covers both sides honestly: what DALL-E 3 was, why it disappeared, and how FLUX compares against it and against the model that took its place.
Most comparison articles online still price DALL-E 3 as if it ships today, which misleads anyone trying to make a real decision in August 2026. This guide corrects that gap with the actual retirement timeline, FLUX’s current FLUX.2 pricing tiers, and GPT Image 2’s specs, sourced from Black Forest Labs’ own pricing page and OpenAI’s product announcements.
By the end, readers know exactly what happened to DALL-E 3 and which tool fits their next image generation project. For a related comparison, see AI Comparison.
DALL-E 3 vs FLUX at a Glance
FLUX remains fully available and actively updated, while DALL-E 3 has been discontinued and replaced by GPT Image 2 as OpenAI’s current image model. The table below summarizes the 2 tools’ status and core specs before the detailed breakdown that follows. For a related comparison, see Best AI Video Generators.
| Category | DALL-E 3 | FLUX |
|---|---|---|
| Developer | OpenAI | Black Forest Labs |
| Status as of August 2026 | Retired May 12, 2026 | Active, current generation FLUX.2 |
| Launched | October 2023 | August 2024 (FLUX.1), FLUX.2 in 2026 |
| Access | No longer available anywhere | API, open-weight downloads, and third-party apps |
| Successor / current option | GPT Image 2 (OpenAI) | Not applicable, FLUX.2 is the current release |
| Max resolution (historical/current) | 1792×1024 | Up to 2048×2048, higher via upscaling |
| Entry API price | $0.04 per image (before retirement) | $0.014 per megapixel (FLUX.2 Klein 4B) |
| Open-weight license | Never offered | Yes, FLUX.2 Dev and Klein tiers |
| Best for | No longer selectable | Photorealism, open-weight self-hosting, text-heavy images |
What Is DALL-E 3?
DALL-E 3 was OpenAI’s third-generation text-to-image model, launched in October 2023 and retired on May 12, 2026. It ran inside ChatGPT and Microsoft’s Bing Image Creator and Designer tools, producing images up to 1792×1024 pixels from natural-language prompts. Its defining strength was semantic prompt understanding: ChatGPT rewrote a user’s short prompt into a longer, more specific one before sending it to the model, which reduced the vague or literal misreads that hurt earlier diffusion models.
OpenAI’s retirement of DALL-E 3 followed a 2-stage transition. ChatGPT quietly swapped its default image model to GPT Image 1.5 in December 2025, and most users never noticed the switch. OpenAI then launched GPT Image 2 on April 21, 2026 under the consumer name ChatGPT Images 2.0, followed by the formal DALL-E 2 and DALL-E 3 API shutdown on May 12, 2026, a date OpenAI had announced back on November 14, 2025.
Nothing running today uses DALL-E 3: any API call referencing the dall-e-3 model ID now returns an error, and ChatGPT, Bing, and Microsoft Designer all route image requests through GPT Image 2 instead. GPT Image 2 keeps DALL-E 3’s default commercial usage rights but adds native support for up to 4K output, character-accurate multilingual text rendering, and an agentic mode that reasons about image structure before generating. Readers evaluating today’s real OpenAI option can compare it directly in the DALL-E 3 vs Adobe Firefly guide, which covers the same transition in more depth.
What Is FLUX?
FLUX is a family of text-to-image and image-editing models built by Black Forest Labs, a German AI lab founded in 2024 by former Stability AI researchers who created the original Stable Diffusion models. FLUX.1 launched in August 2024 across 3 tiers, Schnell, Dev, and Pro, and Black Forest Labs moved to its next-generation FLUX.2 lineup during 2026 with Klein, Dev, Pro, Flex, and Max variants. Unlike DALL-E 3, FLUX ships under a mix of open-weight and commercial API licenses rather than as a single closed product.
FLUX.2’s open-weight tiers, Klein and Dev, can be downloaded and run on a company’s own hardware, which lets developers fine-tune the model on custom datasets and avoid per-image API fees entirely. The API tiers, Pro, Flex, and Max, run on Black Forest Labs’ hosted infrastructure and bill per megapixel of output rather than per fixed image size. FLUX also ships FLUX Kontext, a dedicated image-editing model that changes backgrounds, relights scenes, and swaps objects in an existing photo while preserving the original composition.
Third-party platforms including Freepik, Krea, ComfyUI, Replicate, and fal.ai all offer FLUX access through their own interfaces, so most users never touch Black Forest Labs’ raw API directly. This distribution model is a direct contrast to DALL-E 3’s single-source availability through OpenAI and Microsoft, and it is a large part of why FLUX kept shipping updates while DALL-E 3 was phased out.
Feature Comparison: Prompt Adherence, Text Rendering, and Editing
FLUX’s defining strength is prompt-faithful photorealism, while DALL-E 3’s defining strength was ChatGPT’s automatic prompt rewriting for semantic accuracy. Both approaches solved the same problem, literal versus vague prompts, from opposite directions, and that difference still explains most of the community debate around which tool “wins.”
Prompt Adherence and Semantic Understanding
FLUX follows a prompt’s literal wording closely, rendering exactly what a user describes with comparatively little creative reinterpretation. DALL-E 3 relied on ChatGPT expanding short prompts into longer, more explicit instructions before generation, which produced strong results from vague input but less predictable results from highly specific compositional requests. Community testing found both approaches winning different rounds of the same benchmark, with FLUX excelling on natural-looking scenes and DALL-E 3 excelling on structured, multi-object prompts.
Text Rendering Inside Images
FLUX.1.1 Pro and later FLUX.2 models render legible in-image text accurately, a capability that used to separate DALL-E 3 from most open-weight competitors. DALL-E 3 was also strong at spelled-out text, correctly rendering short phrases, quotes, and headlines baked directly into a generated image. Since GPT Image 2 replaced DALL-E 3, text rendering accuracy has become a race between FLUX.2 and GPT Image 2’s multilingual, character-level text support rather than between FLUX and DALL-E 3.
Editing and Model Access
FLUX Kontext gives creators a dedicated model for background swaps, relighting, and object replacement on existing images, a workflow DALL-E 3 never fully supported on its own. DALL-E 3’s only comparable feature was ChatGPT’s conversational inpainting, which required regenerating an image rather than editing it directly. FLUX’s open-weight Dev and Klein tiers also let a team fine-tune the model for a specific brand style, a level of access DALL-E 3 never offered at any price.
Performance Comparison: Speed and Image Quality
FLUX generates faster at its lower tiers and matches or exceeds DALL-E 3’s historical image quality, particularly on photorealistic subjects. Independent testing throughout 2025 and 2026 measured concrete generation times for both models before DALL-E 3’s retirement, giving a clear speed picture even though one side of the comparison is now historical.
FLUX-Schnell generated an image in roughly 1.3 seconds, FLUX-Pro took 15 to 25 seconds, and DALL-E 3 took 10 to 15 seconds per image during side-by-side tests run before its shutdown. On photorealism specifically, testers across 8 different prompt categories, including portraits, fashion photography, and complex detailed scenes, rated FLUX-Pro’s output as more consistently realistic than DALL-E 3’s, while DALL-E 3 held an edge on stylized, illustrative outputs. FLUX’s material rendering and lighting behavior were repeatedly cited as its strongest technical advantage over both DALL-E 3 and Stable Diffusion in the DALL-E 3 vs Stable Diffusion comparison.
Resolution is another clear gap: FLUX.2 generates natively up to 2048×2048 pixels, compared to DALL-E 3’s maximum of 1792×1024. GPT Image 2, DALL-E 3’s actual successor, now claims native 2K output with experimental support up to 4K, closing that resolution gap on OpenAI’s side.
DALL-E 3 vs FLUX Pricing
FLUX prices its API by megapixel of output starting at $0.014, while DALL-E 3’s pricing is now historical since its API was fully shut down on May 12, 2026. Anyone budgeting for a project today needs FLUX’s current rates and GPT Image 2’s current rates, not DALL-E 3’s old numbers.
| Model / Tier | Price | Status |
|---|---|---|
| DALL-E 3 API (historical) | $0.04 to $0.12 per image | Retired May 12, 2026 |
| ChatGPT Plus (DALL-E 3 era) | $20/month | Now includes GPT Image 2, not DALL-E 3 |
| FLUX.2 Klein 4B | $0.014 per megapixel | Active |
| FLUX.2 Klein 9B | $0.015 per megapixel | Active |
| FLUX.2 Pro | $0.03 per megapixel | Active |
| FLUX.2 Flex | $0.05 per megapixel | Active |
| FLUX.2 Max | $0.07 per megapixel | Active |
| GPT Image 2 (1024×1024, high quality) | $0.211 per image | Active, DALL-E 3’s successor |
FLUX’s open-weight Dev and Klein models can also run on self-hosted hardware, eliminating per-image fees entirely at the cost of covering compute. Black Forest Labs additionally sells fixed licensing tiers, Builder, Platform, Professional, and Enterprise, for teams that need bulk image volume, fine-tuning rights, or multi-domain deployment beyond simple API calls. GPT Image 2 now costs more per image than DALL-E 3 ever did, since its token-based pricing charges $8 per million input tokens and $30 per million output tokens, which scales sharply at higher resolutions.
Pros and Cons
FLUX’s strengths center on photorealism, open-weight flexibility, and megapixel-based pricing, while DALL-E 3’s strengths are now historical since the product no longer exists. Weighing pros and cons here mostly means weighing FLUX against GPT Image 2, not against DALL-E 3 directly.
FLUX Pros and Cons
FLUX’s 4 clearest advantages are strong photorealism and lighting accuracy, sub-2-second generation on its Schnell/Klein tiers, open-weight self-hosting for fine-tuning and cost control, and wide third-party availability through Freepik, Krea, and ComfyUI. Its main drawbacks are a less unified single interface compared to ChatGPT’s all-in-one design and pricing that varies significantly by resolution and tier, which can make budgeting less predictable than a flat per-image rate.
DALL-E 3 Pros and Cons (Historical)
DALL-E 3’s 4 strongest historical advantages were tight ChatGPT integration, strong semantic prompt understanding from automatic prompt rewriting, reliable text-in-image rendering, and built-in commercial usage rights on every paid plan. Its permanent drawback, as of May 12, 2026, is that it no longer exists in any accessible form, making every one of those advantages relevant only to GPT Image 2, the model that inherited them.
User Reviews: What the AI Community Says
Community discussion consistently favors FLUX for photorealism and open-source flexibility, while pre-retirement discussion favored DALL-E 3 for prompt-following through ChatGPT’s rewriting step. Since DALL-E 3’s shutdown, that debate has effectively moved to FLUX versus GPT Image 2.
Users on Reddit and other community forums describe FLUX as state-of-the-art for overall image quality among open and semi-open models, particularly praising its text rendering and natural lighting. Developers who used DALL-E 3 through its final year cited its prompt accuracy for literal, text-heavy generations as its biggest edge, crediting ChatGPT’s prompt-rewriting step for turning vague requests into usable images. A recurring theme in more recent discussion is that FLUX’s open-weight access made it the natural fallback once DALL-E 3 disappeared, since it did not require migrating to a new closed-source OpenAI product.
Use Cases: Which Tool Fits Which Creator
FLUX fits developers, agencies, and creators who need photorealistic output, self-hosting, or fine-tuned brand consistency, while anyone who specifically wants DALL-E 3’s original workflow today uses GPT Image 2 instead. Real-world fit depends on whether a project needs an open, customizable pipeline or a single managed product.
Developers building e-commerce product photography, marketing agencies producing photorealistic ad creative, and technical teams that need on-premises image generation for data privacy all benefit from FLUX’s open-weight Dev and Klein tiers. Solo creators and marketers who want a single conversational interface without managing an API benefit more from GPT Image 2 inside ChatGPT, which now covers the exact workflow DALL-E 3 used to serve. Teams still comparing FLUX against Stable Diffusion’s older open-source lineage before picking a self-hosting stack can see the tradeoffs in the Stable Diffusion vs Flux guide.
Final Recommendation: Choose FLUX or GPT Image 2?
The right pick depends on whether a project needs an open, self-hostable model or a single managed product inside ChatGPT, since DALL-E 3 itself is no longer an option. Choose FLUX if these 4 factors matter most:
- Photorealistic output and accurate lighting matter more than a single unified chat interface
- Open-weight self-hosting or fine-tuning on a custom dataset is required
- Predictable, megapixel-based API pricing fits the budget better than a flat per-image rate
- Access through third-party platforms like Freepik, Krea, or ComfyUI is already part of the workflow
Choose GPT Image 2 if these 4 factors matter most:
- A single conversational interface inside ChatGPT is preferred over managing a separate API
- Native 4K experimental output and character-accurate multilingual text are required
- Agentic reasoning that plans image structure before generating adds real value to the workflow
- Built-in commercial usage rights without a separate licensing tier are a priority
Alternatives to DALL-E 3 and FLUX
Midjourney, Stable Diffusion, and Ideogram are the 3 most-cited alternatives now that DALL-E 3 itself is off the table. Midjourney suits creators who want the most stylistically distinctive output, a comparison covered in the Midjourney vs Flux guide. Stable Diffusion remains the most customizable fully open-source option, while Ideogram focuses specifically on typography and logo-style text generation.
For a complete breakdown of the top AI image generation platforms in 2026, the Best AI Image Generators guide compares multiple tools across pricing, quality, and workflow fit. Creators evaluating OpenAI’s current lineup against Adobe’s competing model can also see how GPT Image 2 stacks up in the DALL-E 3 vs Adobe Firefly comparison.
Frequently Asked Questions
Is DALL-E 3 still available in 2026?
No, DALL-E 3 is not available anywhere as of August 2026. OpenAI retired both the DALL-E 2 and DALL-E 3 APIs on May 12, 2026, and ChatGPT, Bing Image Creator, and Microsoft Designer all now generate images through GPT Image 2 instead.
What replaced DALL-E 3?
GPT Image 2 replaced DALL-E 3, launching April 21, 2026 under the consumer name ChatGPT Images 2.0. ChatGPT had already quietly switched its default image model to the interim GPT Image 1.5 in December 2025 before GPT Image 2’s full release.
Is FLUX better than DALL-E 3?
FLUX generally outperforms DALL-E 3 on photorealism, lighting accuracy, and generation speed at its faster tiers, based on side-by-side testing conducted before DALL-E 3’s retirement. DALL-E 3 held an edge on literal prompt-following through ChatGPT’s automatic prompt rewriting, a strength GPT Image 2 has since inherited.
Is FLUX free to use?
FLUX offers free access through limited daily credits on some third-party platforms like Krea and Freepik, and its Klein and Dev models can be downloaded and self-hosted under an open-weight license. Black Forest Labs’ own hosted API, by contrast, charges per megapixel starting at $0.014 for FLUX.2 Klein 4B, with no permanent free tier on the official API itself.
Can I still access images I generated with DALL-E 3?
Yes, previously generated DALL-E 3 images remain available to users who already created them, even though the ability to generate new images through DALL-E 3 ended on May 12, 2026. New generation requests through ChatGPT, the API, or Microsoft’s tools now route to GPT Image 2 automatically.
Which is better for text rendering, FLUX or DALL-E 3?
FLUX.1.1 Pro and later FLUX.2 models render accurate, legible in-image text, a capability that once set DALL-E 3 apart from most competitors. Since DALL-E 3’s retirement, the more relevant comparison is FLUX against GPT Image 2, which adds character-level accuracy for multilingual, non-Latin text that DALL-E 3 never supported.
Final Verdict
FLUX is the actively developed choice between the two, while DALL-E 3 is a discontinued product whose real successor, GPT Image 2, is what anyone searching this comparison should actually evaluate. FLUX wins for developers and agencies who need photorealistic output, open-weight self-hosting, or predictable megapixel-based pricing across third-party platforms like Freepik and ComfyUI. GPT Image 2 wins for creators who want DALL-E 3’s original conversational, all-in-one ChatGPT workflow, now upgraded with native 4K experimental output and agentic image reasoning.
Anyone still budgeting around DALL-E 3’s old $0.04 to $0.12 per-image API pricing gets more accurate numbers from FLUX’s current $0.014 to $0.07 per-megapixel tiers or GPT Image 2’s $0.211-per-image starting rate. The deciding factor is not which tool has the longer feature list, but whether a project needs an open, self-hostable model or a single managed product inside ChatGPT.
