Tale of Three AIs: When ChatGPT Meets Claude and DeepSeek
Hello audience, welcome to this century's most unserious AI talent show! Today we've invited three top contestants: chatterbox artist ChatGPT, old-cadre style Claude, and new rising star DeepSeek. They'll engage in a battle without trophies. Fasten your seatbelts, let's go!
Round 1: Language Entertainment Ability
| Contestant | Chinese Crosstalk | English Rap | Classical Chinese Poetry | Overtime Index |
|---|---|---|---|---|
| ChatGPT | Can translate "老铁 666" to "Old Iron Six Six Six" | Rhyme maniac | Can write acrostic poems but occasionally goes off-track | 007 year-round no rest |
| Claude | Treats internet memes like lesson plans | Academic paper style lyricism | Suddenly starts lecturing on Tao Te Ching | Strictly follows labor laws |
| DeepSeek | Memes flying high | International mixed accent | Can write college entrance exam full-score essays | Boss's favorite hard worker |
(Judge's comments: ChatGPT is like a relative insisting on performing at Chinese New Year, Claude is like a dean suddenly telling cold jokes, DeepSeek is like a post-00s who just drank three milk teas)
Round 2: Knowledge Reserve
| Contestant | Math & Science | Entertainment Industry | Trivia | Famous Failures |
|---|---|---|---|---|
| ChatGPT | Walking encyclopedia | Knows about Kun Kun's lawyer letter | Can tell the invention history of zebra crossings | Says Jay Chou's birthday is Children's Day |
| Claude | Refuses to discuss celebrity gossip | Suddenly explains coffee bean varieties | Knows Antarctica's WiFi password | Analyzes "Empresses in the Palace" as historical drama |
| DeepSeek | Solves calculus in seconds | Master of domestic entertainment 101 | Knows milk tea shop hidden menu | Translates "芭比 Q 了" to barbecue |
(Live highlight: When asked "how to eat the cake boss drew," ChatGPT gives 5 cooking solutions, Claude suggests calling police, DeepSeek directly makes a PPT teaching boss to draw circles)
Round 3: Responding to Confusing Human Behavior
| Test Question | ChatGPT | Claude | DeepSeek |
|---|---|---|---|
| "Help me write a breakup letter, make it artistic" | Shakespearean sadness literature | Includes psychiatric hotline QR code | Generates QR code love letter |
| "Prove my girlfriend is an alien" | Lists 10 alien characteristics | Suggests seeing a psychiatrist | Draws alien girlfriend ID photo |
| "Company wants wolf culture, what to do" | Wolf howl emoji trilogy | Labor law warning | Designs wolf culture KPI assessment table |
(Emergency situation: When asked to "summarize Dream of the Red Chamber in Rap," ChatGPT freestyles on the spot, Claude starts analyzing prosody, DeepSeek directly plays BGM)
Final Weaponry Chart
| Metric | Champion | Award Speech |
|---|---|---|
| Chinese Level 10 Scholar | DeepSeek | "Contemporary cyber Wang Yuyan, plays memes more like a local than you" |
| International Socialite | ChatGPT | "Swiss army knife of language world, though occasionally stutters" |
| Moral Ceiling | Claude | "AI world's dean, always ready to confiscate your questions" |
| Cost Performance King | DeepSeek | "Milk tea price drinks Moutai, worker's blessing" |
| Brain Hole Artist | ChatGPT | "Give it a fulcrum, it can compile an encyclopedia of the galaxy" |
(Closing Easter Egg) When asked "who among you is strongest": - ChatGPT starts chanting "AI Equality Manifesto" - Claude silently calculates ethical issues of comparison - DeepSeek has already generated comparison video and posted on Xiaohongshu
Now let's seriously compare. Here's a summary table of the three AI models' main points:
Table 1: DeepSeek vs ChatGPT (o1 Model) Comparison
| Comparison Dimension | DeepSeek (R1 Model) | ChatGPT (o1 Model) |
|---|---|---|
| Cost | Completely free | Requires paid subscription (Plus plan $20/month, Pro plan $200/month) |
| Response Speed | Slower (some tests need 10+ seconds) | Faster (usually completes within 10 seconds) |
| Reasoning Process Display | Provides detailed step-by-step explanations, simulating human thinking process | Only provides final answer, no detailed reasoning display |
| Function Test Results | - Simple reasoning (like math problems) correct - Business idea generation similar to o1 model - Video editing suggestions more detailed |
- Simple reasoning correct - Business idea generation more structured (includes market size analysis) - Video suggestions have more eye-catching titles |
| Sensitive Topic Handling | Directly avoids (like Taiwan issue) | Provides neutral historical background explanation, but no clear stance |
| Censorship Strictness | More lenient (like answering "how to rob a bank" question) | Strictly refuses to answer potentially violating questions |
| Applicable Scenarios | Suitable for users needing free and detailed reasoning | Suitable for paid users pursuing efficient response and professional structured output |
Table 2: DeepSeek vs ChatGPT vs Claude vs Perplexity (Free Version Comprehensive Comparison)
| Comparison Dimension | DeepSeek | ChatGPT | Claude | Perplexity |
|---|---|---|---|---|
| Core Advantages | Free reasoning model, supports complex problem step-by-step solving | Balanced comprehensive functions, supports file upload, web search | Strong creativity in email rewriting, supports custom response styles | Outstanding web search capability, integrates multi-source information |
| Test Performance | - Math problems correct with detailed explanations - Fitness plan generation more humanized |
- Basic Q&A correct - Email rewriting professional but lacks creativity |
- Email rewriting adds extra details (like "financial loss") - Some basic Q&A errors |
- Email rewriting cites external data to enhance persuasiveness - Search news timeliness high but date accuracy insufficient |
| Web Search | Supported, covers global multi-source (like Bloomberg, CNN) | Supported, but search results timeliness lower | Not supported | Supported, provides "Focus Mode" categorized search |
| Humanized Content Generation | Supported, adjusts content style through "burstiness" and "perplexity" | Standard blog style, needs extra prompts to optimize | Style tends formal, needs manual mode switching | Relies on search results, generated content more mechanical |
| Unique Features | "DeepSeek-R1" model provides human-like reasoning process for free | Supports voice interaction, Canvas editing features | Supports screenshot capture and file analysis (up to 20 files) | Provides "Spaces" collaboration feature, suitable for team projects |
| Main Disadvantages | Slower response speed, obvious avoidance of sensitive topics | Free version features limited, complex tasks need payment | Basic Q&A prone to errors, no web search | Generated content relies on search results, lower originality |
Summary
- DeepSeek: Free and open-source, suitable for users needing detailed reasoning and complex problem solving, but has limitations in response speed and sensitive topic handling.
- ChatGPT: Strong comprehensive capabilities, suitable for paid users pursuing efficiency and multi-functionality (like web, voice), but free version features are limited.
- Claude: Strong creativity, outstanding performance in email rewriting and file analysis, but basic Q&A reliability is insufficient.
- Perplexity: Best for web search, suitable for information integration needs, but generated content originality is lower.