MandAPIQwen4 preview: Max, Plus, Flash, 27B and what Qoduck really offers
Updated November 8, 2026. Alibaba has not published Qwen4 specifications, pricing or a release date.
Alibaba says it is working on the next Qwen generation. The names Qwen4-Max, Qwen4-Plus, Qwen4-Flash and Qwen4-27B have circulated through reports and images of conference slides. Names alone do not tell us the models' speed, context window, license or hardware needs. Separately, Qoder now offers Qoduck, an undisclosed model with a daily free allowance. Qoder does not say that Qoduck is Qwen4.
Has Qwen4 been released?
As of November 8, the official sources we checked do not provide Qwen4 weights, a model card or an API price sheet. Alibaba Cloud's September announcement described Qwen4 as being trained. Its projection of 5–10 trillion parameters applies to future Qwen4.5 and Qwen5 series, not to Qwen4 itself.
The four tier names appear in reporting about conference slides, rather than in the technical specifications of that announcement. Treat them as a provisional map:
| Name in circulation | Plausible role | Still unknown |
|---|---|---|
| Qwen4-Max | Higher-capability tier | Price, benchmarks, availability |
| Qwen4-Plus | Balanced tier | Limits, latency, price |
| Qwen4-Flash | Efficiency tier | Throughput, token cost |
| Qwen4-27B | Smaller model | License, weights, quantization, memory |
Those roles are inferences from names, not Alibaba performance commitments.
Will Qwen4-27B run on my PC?
The 27B label is interesting for local inference, but it does not promise that a 16 GB or 24 GB GPU will work. Memory depends on the released weights, quantization, context length, KV cache and runtime. Wait for an actual model card and license before buying hardware. Alibaba has discussed running existing Qwen3.8 27B models on consumer hardware; that is useful background, not a Qwen4-27B requirement.
Is Qoduck Qwen4? How does the free offer work?
The official Qoder offer calls Qoduck a stealth model whose identity has not been disclosed. Claims that it is “Qwen4-Max Preview” remain speculation. To try it, update the main Qoder International app, sign in with an individual account and select Qoduck. Free accounts and accounts with zero Credits are eligible for a daily allowance.
Within that allowance, calls cost zero Credits, even at the highest thinking intensity. Once it is exhausted, additional use costs the standard 0.5× Credits, or you can wait for the next daily reset. Qoder IDE, JetBrains, CLI, Teams and Enterprise are excluded. Qoder has not announced an end date.
What should developers do now?
For coding, test Qoduck on your own tasks and measure quality, latency and usage. For API integration, wait for an official Qwen4 model identifier and documentation. The Qoder offer is an application feature, not a public Qwen4 API.
For work today, browse the MandAPI model selection guide and check current prices. The guide helps compare task fit, context and cost. This article does not claim MandAPI access to Qwen4 or Qoduck.
Frequently asked questions
When will Qwen4 launch?
Alibaba had not published an official launch date in the sources checked on November 8, 2026.
Is Qoduck Qwen4-Max Preview?
That has not been confirmed. Qoder describes Qoduck only as a stealth model with an undisclosed identity.
Is Qoduck still free after the daily allowance runs out?
No. Further use consumes Credits at the 0.5× rate; the free allowance resets the next day.
Should I buy a GPU for Qwen4-27B now?
Plan scenarios, but wait for weights, license, quantization and memory requirements before buying hardware.