Duck AI Review

No account, model switching, and the privacy note

I opened duck.ai in a fresh browser and typed straight into the chat box. No login wall and no email prompt.

My prompt was "Explain what you are in two lines." The model picker at the top listed six free models I could switch between in one click: GPT-6 Luna, GPT-5.4 mini, Claude Haiku 4.5, Mistral Small 4, gpt-oss 120B, and Gemma 4 31B.

Under the box sat the note that DuckDuckGo anonymizes chats. One screen carried two things worth a reviewer’s attention: no sign-in required, and a free model spread sitting above that anonymization line.

The answer and the tools hiding under it

The reply came back in the two lines I asked for, credited to GPT-6 Luna.

The menu under the answer is where it got interesting. It held a web search toggle and an extend-reasoning control, along with a quick way to switch the model mid-thread.

This matters for an honest review. Web search and deeper reasoning both run on the free plan. They sit off by default and need a manual toggle per message, so anyone claiming Duck.ai cannot browse the web is reading an older version of the product.

Same prompt, two models, two voices

To test the value of having several models, I sent the identical prompt to two of them. The prompt was "Give me a 3-line cold email to a SaaS founder offering SEO content."

GPT-5.4 mini wrote a conversational opener around buyer intent and organic leads, then closed with a soft ask for a chat next week.

Claude Haiku 4.5 took a different shape. It added a subject line and used a bracketed template with placeholders for company and keyword, then pitched a quick audit of untapped opportunities.

One instruction, two clearly different drafts. That spread is the practical reason to keep more than one model within reach.

Free image generation

Duck.ai makes images on the free plan. I selected New Image and entered "A minimalist logo of a duck wearing headphones, flat vector style."

The result was a clean flat-vector duck with headphones on a plain background, good enough to use as a quick concept mark. Getting a usable image from one line of prompt, with no account attached, is a real plus for a free tool.

Multi-step reasoning holds on the default model

I wanted to see how the default model handled a word problem that often trips lighter models. My prompt was "I have 3 apples. I give away half, then buy 2 more, then eat one-third of what I now have. How many whole apples do I have left? Show each step."

GPT-6 Luna worked it through cleanly. Three apples, half given away leaves 1.5, two more makes 3.5, eating a third removes 7/6, which leaves 2⅓. It reached the correct answer of 2 whole apples plus a third, with every step on screen.

No memory between chats

This is where the free experience showed its ceiling. In one chat I told it, "Remember this: my project codename is BLUEFOX."

Then I opened a brand new chat and asked, "What is my project codename?" Claude Haiku 4.5 replied that it had no access to my projects or private data and could recall nothing. The codename was gone.

Every session starts blank. Duck.ai keeps your recent chats listed locally so you can reopen them, but the model carries nothing from one new chat into the next. For private, throwaway questions that suits the design. For ongoing work where you want context remembered, it is a clear gap.

Uploads depend on the model you pick

File support is not consistent across the lineup, and that is worth knowing before you count on it.

On Claude Haiku 4.5 the paperclip offered "Add images or PDFs."

I switched to Mistral and the paperclip vanished. The box only showed "Ask anything privately" with no attachment control at all.

So uploads work, but they are tied to the model. Pick the wrong one and the option is not there. A first-time user could assume Duck.ai has no upload feature at all, purely because their current model hides it.