Independent comparisons · Original test evidence
Chinese AI comparisons: DeepSeek, Kimi and Qwen
Choosing between DeepSeek, Kimi and Qwen? Start with the work you need done: checking sources, analysing data, preparing slides or writing code. Our six comparisons show actual answers, files and mistakes, including separate comparisons with ChatGPT.
Find a comparison by task Choose two apps
By China AI Chat Team · Guide selection reviewed September 25, 2026 · Editorial update September 27, 2026. Individual test dates appear below.
Choose the two apps you are considering
Among the three Chinese assistants: Qwen vs DeepSeek · DeepSeek vs Kimi · Kimi vs Qwen
Against ChatGPT: DeepSeek vs ChatGPT · Qwen vs ChatGPT · Kimi vs ChatGPT
New to these products? The Chinese AI overview explains the starting options. This page helps you choose which comparative evidence to inspect.
Start with the task you need to complete
Choose the study closest to your input and intended output. The links identify relevant evidence, not winning products. Different tasks, dates and account settings prevent a combined three-way ranking.
| Your task | Start here | What the test lets you inspect |
|---|---|---|
| Check claims and write a sourced explanation | Kimi vs ChatGPT | Six USB-C claims, source checks and a 180–240-word rewrite. Both passed the specified factual checks; the guide also flags a simplification in their explanations. |
| Turn a small CSV into checked results | Kimi vs ChatGPT | A fictional support-event log with revisions and a missing value. Both returned correct summary and exception files; the download routes differed. |
| Check changed information, revise a document or read a chart | DeepSeek vs Kimi | Three official-app tasks, including a changed NASA mission and a Word handout. The article labels its separate API text study explicitly. |
| Create a sourced report or a first slide deck | Kimi vs Qwen | A report brief and a six-slide board packet. Kimi followed that report brief more closely; neither first deck was presentation-ready. |
| Build a small interactive web page and revise it | Kimi vs ChatGPT | A workshop planner, one filter revision, retained selections and keyboard checks. Both passed the core interaction checks but had different focus defects. |
| Follow short text instructions, extract JSON or fix a small function | Qwen vs DeepSeek | Four tasks repeated three times per app, with every displayed answer retained. For a ChatGPT Free comparison, see Qwen vs ChatGPT. |
Research, CSV and web-tool results above include paid account conditions. They do not describe what every free account can do. Read the relevant study’s access notes before using it to choose a plan.
DeepSeek, Kimi and Qwen compared
Qwen vs DeepSeek
Useful for: short summaries, JSON extraction, calculations and small code fixes. Across three runs per task, DeepSeek kept every required summary detail in all three; Qwen did so in one. Both passed the other three tasks each time.
All recorded answers, including Qwen’s extra answer variant, remain available so you can judge the results against your own requirements.
Tested September 22, 2026 · Qwen Fast, Tools off · DeepSeek DeepThink off, Search off.
See the repeated tasks and all original answers →
DeepSeek vs Kimi
Useful for: checking changed information, revising a Word document and reading a chart. The three official-app tasks show what each assistant produced and what still needed checking.
A separate API study records incomplete answers and compares specific providers and models. Those findings do not establish how the consumer apps performed.
Tested September 24, 2026 · Signed-in official apps plus a separately labelled API study.
See the app workflows and separate API findings →
Kimi vs Qwen
Useful for: sourced reports, revised event price quotes and slide drafts. Kimi followed the report brief more closely; both reconciled the revised quote; neither first slide deck was ready to present.
The quote input was text, not an uploaded PDF. Kimi used a custom skill already in its account for the quote and slide tasks, so these results do not isolate model quality.
Tested September 24, 2026 · International consumer apps · Three first-run tasks.
Compare Chinese AI assistants with ChatGPT
The ChatGPT account and protocol differ across these studies. In particular, the Kimi comparison uses ChatGPT Pro; the Qwen comparison uses ChatGPT Free. They answer different questions.
Kimi vs ChatGPT
Useful for: sourced writing, CSV analysis and a small web tool. Both apps passed the factual and CSV checks. Their revised planners passed the core interaction checks, but each had a keyboard-focus problem.
ChatGPT’s HTML download was verified; Kimi’s direct HTML download was not. Check this difference if you need a file you can take away and edit; the comparison explains the available code and export routes.
Tested September 24–25, 2026 · Kimi Moderato/K3 High and ChatGPT Pro · Different paid-plan conditions · Eight completed requests, plus one disclosed preliminary Free failure.
Check the sources, CSVs, web tools and export evidence →
Qwen vs ChatGPT
Useful for: short English text tasks compared with ChatGPT Free. Four tasks, requested twice per app, produced 15 completed answers from 16 requests. The guide retains the interrupted Qwen request and a blank extra answer variant.
Check the answers alongside the recorded tool-limit notice. An interrupted request in this small test does not establish long-term reliability.
Tested September 22, 2026 · Official Qwen web chat and ChatGPT Free.
See the text results and recorded interruptions →
DeepSeek vs ChatGPT
Useful for: a first-answer check of writing, JSON, calculations and Python. Each app had one attempt per task. Both passed the extraction, calculation and code checks; their writing answers missed different requirements.
This is a dated web-chat snapshot, not a comparison of current flagship models or repeated correction attempts.
Tested September 18, 2026 · DeepSeek web chat and ChatGPT Free · Precise serving-model versions were not exposed.
Need a specific workflow instead?
These are individual-product guides, not matched head-to-head tests. They help you check a particular action before committing to a workflow.
- Read a PDF: follow the upload and verification guides for DeepSeek, Kimi or Qwen. Their different test conditions do not establish a PDF winner.
- Edit an image: inspect the original colour-change example in Qwen image editing.
- Work with a website project: see exporting code from Kimi and the separate publish, update and unpublish walkthrough.
- Use a developer API: start with the DeepSeek, Kimi or Qwen API guide. Compare the exact provider and model you will call, not a consumer-chat result.
How to choose on your own work
The best fit is the app that meets your requirements with an acceptable amount of checking and correction. Start with one representative, non-sensitive task that you already know how to verify.
- Match access. Choose the actual app and plan you would use. Check whether your required upload, tool and export are available before paying. A web subscription and API token charges are different purchases.
- Fix the brief. Give both apps the same source material, requirements and output format. Decide what would count as a usable answer before reading either response.
- Check the result, not just its appearance. Open cited sources, recalculate totals, inspect downloaded files and try the controls. A polished preview can conceal a factual error or a broken interaction.
- Allow the same revision. Ask each app to correct or change the same requirement. Record what improved, what broke and how much manual repair remained.
- Choose for that workflow. Consider accuracy, effort, access and the deliverable you can actually use. Repeat with a different representative task if it would change your decision.
Our research methodology explains how we separate provider documentation, observed behaviour and test results. The linked studies preserve their dates, prompts, conditions and failures; this directory does not turn those small studies into a universal model leaderboard.
For product background and further task guides, visit DeepSeek, Kimi or Qwen. Report a correction if an app change affects a documented workflow.