Qwen or DeepSeek via API: Price, Context and Intelligence Index

When you need a cheap model for bulk work, the choice often comes down to two Chinese lineups. Below is a Qwen API vs DeepSeek comparison: Qwen 3.8 Max and Flash against DeepSeek V4 Pro and V4.1 Flash. We compare only what a table can confirm: price, context, image support, intelligence index, and the cost of one typical task. The question of which one writes better stays with your own test on real data.

Four models in one table

Prices as of October 18, 2026, $ per 1M tokens. "Index" is the Intelligence Index from Artificial Analysis.

What stands out:

• All four have a 1M-token context, so document length does not separate them.
• Only Qwen 3.8 Max and DeepSeek V4 Pro accept image input. If your task has screenshots or scans, the Flash versions will not do.
• On the index, Qwen 3.8 Max is highest (45), V4.1 Flash and Qwen Flash are close (39 and 40), and V4 Pro scores 36.
• The smaller DeepSeek model is clearly the cheapest on both input and output.

The index is an averaged score on common tests. It says the models are broadly comparable in class, but not how each handles your emails or your code.

Pricing one task

Take an illustrative document-analysis request: 20,000 input tokens, 1000 output. Formula: input / 1M × input price + output / 1M × output price.

The gap between the most and least expensive model here is about $1.70 per thousand tasks. At small volumes that is pennies, so it makes sense to pick the model with the best result on your data. At tens of millions of requests the difference becomes a real line item.

What about cache

In the Zukko table the cache-read price for all four models equals the input price: 0.088 for Qwen 3.8 Max, 0.038 for Qwen Flash, 0.075 for V4 Pro, 0.013 for V4.1 Flash. That means a repeated prefix (a long system prompt, a knowledge base at the start of the request) gets no separate discount here, unlike Claude or GPT. You do not need to estimate cache savings for these lineups in advance: count at the plain input price. How cache works for other models is covered in the article on prompt caching→.

What about Kimi and GLM

People often compare other Chinese lineups in the same query. Kimi models are not in the Zukko list as of October 18, 2026, so they cannot be compared on gateway rates. GLM is available:

The same task with 20,000 input and 1000 output tokens: glm-5.3 is 20,000 × 0.1 / 1M + 1000 × 0.1 / 1M = 0.002 + 0.0001 = $0.0021; glm-5.3-flash is 0.0008 + 0.00004 = $0.00084. GLM's output price equals its input price, so it is relatively cheaper for long generations, but the 128K context has to be taken into account. More in the article Qwen versus GLM→.

How to choose

1. Do you need image input? Then Qwen 3.8 Max and DeepSeek V4 Pro remain.
2. Do you need the highest index of the four? By the table that is Qwen 3.8 Max, but it is also the most expensive of the group.
3. Do you need the lowest price for bulk text tasks? Look at V4.1 Flash and Qwen 3.8 Flash.
4. Documents longer than 128K? GLM drops out; any of the four works.

Then take 30-50 real examples, run them through two or three models and compare the answers blind. A simple way to organize this is in the article on testing prompts→.

Switching between them

All four models are available through one address and one key, so comparison comes down to changing the model string:

import os
from openai import OpenAI

client = OpenAI(base_url="https://api.zukko.pro/v1", api_key=os.environ["ZUKKO_API_KEY"])

for model in ["qwen3.8-max", "qwen3.8-flash", "deepseek-v4-pro", "deepseek-v4.1-flash"]:
    r = client.chat.completions.create(
        model=model,
        messages=[{"role": "user", "content": "Briefly list pros and cons of microservices."}],
        max_tokens=300,
    )
    print(model, r.usage.prompt_tokens, r.usage.completion_tokens)
    print(r.choices[0].message.content, "\n")

The current list of models and prices is GET https://api.zukko.pro/v1/models or the Models page in the dashboard. If a model is temporarily switched off, the request returns 503 model_disabled: switch to another one from the list.

FAQ

Which is cheaper, Qwen or DeepSeek?

By the table as of October 18, 2026 the cheapest of the four is DeepSeek V4.1 Flash. Qwen 3.8 Flash is cheaper than DeepSeek V4 Pro, while Qwen 3.8 Max costs more than V4 Pro. Compare specific pairs.

Do Qwen and DeepSeek give a cache discount?

In the table the cache-read price for both lineups equals the input price, so there is no separate discount. Calculate at the plain input price.

Can I use both models in one application?

Yes, the key and address are shared. A common scheme: a cheap model handles the main flow, and hard cases go to a stronger one. See the article on model cascades→.

Can I get access from Russia?

Yes, through the gateway. Payment: SBP, MasterCard or crypto, top-ups from $1.

To compare models on your own data, create a key in the Zukko dashboard and run the code above.