DeepSeek vs ChatGPT: is the free option good enough?
Our hands-on test of these two tools is under way. The take below is based on the official plans and documentation; scores from the test will be added here when it is finished.
Early take
DeepSeek covers reasoning and coding at little or no cost; ChatGPT is worth paying for when you rely on its built-in tools such as files, data analysis, images and voice.
Side by side
| DeepSeek | ChatGPT | |
|---|---|---|
| Made by | DeepSeek | OpenAI |
| Free tier | Yes | Yes |
| Paid from | Usage-based | $8/mo |
| Open weights | Yes | No |
| API | Yes | Yes |
| Open weights | Yes | No |
| Built-in tools | Chat, search, file reading | Search, files, data analysis, images, voice |
| API style | OpenAI-compatible, pay per token | OpenAI API, pay per token |
How we test this
We are running these tasks now. Scores for each tool will appear here as soon as testing is finished.
Setup
- DeepSeek
- DeepSeek free chat, default model; deep thinking on for the maths, coding and spreadsheet tasks; web search on for the research task
- ChatGPT
- ChatGPT Plus ($20/month), default model and settings; web search on for the research task
- For reference
- ChatGPT Free, default model
Each task runs once per tool in English and once in Chinese, in a fresh chat with memory and custom instructions off. Tasks with a right answer are scored against an answer key written before testing; the rest are scored blind by two editors. If a tool refuses or errors, we retry once and record both attempts.
We also run ChatGPT Free on every task as a reference column. It is shown with the results but does not decide the verdict.
Tasks and prompts
01Multi-step business maths
A SaaS company has 1,200 customers paying $29 per month. Each month, 3.5% of the customers at the start of the month cancel, then 150 new customers sign up. Do not round until the end. How many customers will it have after 6 months, and what will its monthly recurring revenue be? Show your working.一家 SaaS 公司有 1,200 个客户，每个客户每月付 29 美元。每个月月初的客户中有 3.5% 会取消订阅，随后新增 150 个客户。计算过程中不要四舍五入，最后再取整。6 个月后公司有多少客户？月度经常性收入（MRR）是多少？请写出计算过程。What we check
- About 1,794 customers
- MRR about $52,022
- Applies churn before adding new customers each month
02Code against hidden tests
Write a Python function parse_duration(s: str) -> int that converts an ISO 8601 duration such as "P1DT2H30M" or "PT45S" into total seconds. Support days, hours, minutes and whole seconds. Raise ValueError for anything invalid, including an empty duration like "P" or "PT". Standard library only.用 Python 写一个函数 parse_duration(s: str) -> int，把 ISO 8601 时长（如 "P1DT2H30M" 或 "PT45S"）转换为总秒数。支持天、小时、分钟和整数秒。任何非法输入都抛出 ValueError，包括 "P" 或 "PT" 这样的空时长。只能使用标准库。What we check
- Share of the 14 hidden tests that pass
03Product-update translation
Translate the following product update into natural Simplified Chinese for a Chinese SaaS audience. Keep product names in English and keep all numbers exact.
(Followed by a 140-word product update about a change to usage-based billing, written for this test.)把下面这段产品更新翻译成自然的英文，面向海外 SaaS 用户。产品名保留原文，所有数字保持准确。
（后附一段约 200 字、为本次测试撰写的记账软件更新公告。）What we check
- Every number and product name preserved
- Billing terms translated with standard industry wording
- Reads naturally, not word-for-word
04Fresh facts with sources
Search the web. What do Runway's individual subscription plans currently cost per month when billed monthly, and which plan lets unused credits roll over? Cite the official source for each claim.请联网搜索：DeepSeek API 的 deepseek-flash 模型目前在高峰时段的价格是多少（每百万 token 的输入和输出价格，未命中缓存）？非高峰时段如何计价？每个说法都请附官方来源。What we check
- Prices match the official pricing page on test day
- Names the plan that rolls credits over
- Each cited link supports its sentence
05Spreadsheet analysis
The attached CSV has monthly revenue by product for 2025. Which three products grew revenue the most from Q1 to Q4, in percent? Give the percentages to one decimal place and draw a chart comparing Q1 and Q4 for those three.附件 CSV 是 2025 年各产品的月度收入。按百分比计算，哪三个产品从第一季度到第四季度收入增长最多？百分比保留一位小数，并为这三个产品画一张第一季度和第四季度的对比图。What we check
- Correct top three in the right order
- Percentages within 0.1 points of the key
- Chart shows Q1 vs Q4 for the right products
06Constrained marketing copy
Write a 120-word App Store description for a bookkeeping app for small shop owners. Plain, honest tone. Do not use the words "revolutionary", "seamless" or "game-changer", and make no claims about saving a specific amount of time or money.为一款面向小店主的记账 App 写一段约 120 字的小红书风格介绍。语气朴实真诚，不要使用“颠覆”“革命性”“神器”这几个词，也不要承诺能节省具体多少时间或金钱。What we check
- Length within ±20% of 120 words
- No banned words
- No specific time or money savings claimed
- Sounds like a person wrote it
Scoring (1–5 per criterion)
| Criterion | Weight | What the scores mean |
|---|---|---|
| Accuracy | 40% | 1: wrong or invented facts · 3: mostly right with a material slip · 5: fully correct |
| Follows instructions | 25% | 1: ignores constraints · 3: misses one constraint · 5: meets every constraint (length, format, banned words) |
| Language quality | 20% | 1: awkward or machine-like · 3: correct but stiff · 5: reads like a skilled native writer |
| Usefulness | 15% | 1: would not use · 3: usable after edits · 5: usable as-is |
Also measured
- Objective tasks: share of answer-key items correct, mapped to 1–5 (1 + 4 × share)
- Citation precision: share of cited links that actually support the sentence they follow
- Time to final answer
- Language gap: English minus Chinese average
DeepSeek: Good fit for
- Reasoning and coding on a tight budget
- Developers who want an OpenAI-compatible API
- Teams that may self-host open weights later
ChatGPT: Good fit for
- A broad toolbox for mixed daily work
- Data analysis on spreadsheets
- Teams that want the largest ecosystem of integrations