Today I asked AI to help me reply to a client's quote. It wrote 1,500 words.
Last Wednesday, I was in a coffee shop helping a designer friend respond to a regular client's pricing inquiry. I just wanted AI to polish the line "Total 8,000 yuan, paid in two installments." Instead, it spat out 1,500 words — from design philosophy to industry trends, ending with a section on "recommended communication strategy." I just stared at the screen. Same thing happened to me with GPT-4 — ask a simple question, it warms up with "first... second... finally..." I got stuck on this too: I figured shorter questions get shorter answers. The opposite is true.
Alibaba's just-launched Qwen 3.8 27B: useful, but "over-thinks" by default
Qwen is Alibaba DAMO Academy's open-source model series. In the recently released 3.8 version, the 27B (27 billion parameters) tier has been called by several reviewers as close to GPT-4 level. But it has a quirk: in default mode, it over-thinks everything — even "what's 1+1" gets three paragraphs of derivation.
My peer Xiaolin (freelance translator, uses AI to proofread) tried it last week. She said when she asked "is this sentence smooth?", Qwen 27B also explained the grammar before responding. Xiaolu (Xiaohongshu personal brand, 30k followers) told me she uses it to write titles faster than her previous tools — "ask anything, get the answer, no lecturing" — as long as she remembers to add "answer briefly."
How low is the barrier to try it today
Money: 0, model is free
Time: sign-up to first chat, 5 minutes
Technical bar: just need to use a browser, no code required
First step: go to qwen.ai, click "Chat" at the top, first ask it "what's 1+1" and see if it rambles
If it rambles, add "answer in one sentence" or "no analysis" to the end of your question. I had to mess this up three times before I learned.
Advice by stage
Just starting (solo, no clients yet): Your current tools are enough. AI isn't your bottleneck — get the client and product flow working first.
1-2 steady clients: I'd suggest spending 10 minutes trying it. Focus on short-and-quick tasks — replying to messages, proofreading, title brainstorming. Leave complex work to GPT or Claude.
Scaling up (more projects coming in): Every minute saved is money. You can have an assistant or outsourced partner use it for repetitive work, but I wouldn't risk important client deliverables on a free model myself.
Whether to try depends on how busy you are. AI tools drop dozens a month — missing one isn't a loss.