DeepSeek V4 Pro
VS
GPT-5 Mini
DeepSeek V4 Pro vs GPT-5 Mini: which wins at real work?
22 task areas · same graded test runs · rank comparison only, so 0–100 and Elo collections never mix raw scores.
DeepSeek V4 Pro wins 16 of 22 task areas we tested; GPT-5 Mini takes 6. DeepSeek V4 Pro costs 1.7× less per token ($1.305 vs $2.25 per 1M).
DeepSeek V4 Pro costs 1.7× less per token ($1.305 vs $2.25 per 1M).
Task by task
| Task area | DeepSeek V4 Pro | GPT-5 Mini | Winner |
|---|---|---|---|
| Executive Assistant |
#38
/ 112
Strong
|
#108
/ 112
Usable
|
DeepSeek V4 Pro |
| Sales |
#44
/ 110
Usable
|
#102
/ 110
Needs editing
|
DeepSeek V4 Pro |
| Customer Support |
#52
/ 113
Strong
|
#104
/ 113
Usable
|
DeepSeek V4 Pro |
| RAG, Safety & Grounding |
#27
/ 113
Excellent
|
#78
/ 113
Excellent
|
DeepSeek V4 Pro |
| Content & Brand |
#29
/ 124
Strong
|
#69
/ 124
Strong
|
DeepSeek V4 Pro |
| Training & Education |
#41
/ 110
Excellent
|
#81
/ 110
Strong
|
DeepSeek V4 Pro |
| Landing Pages |
#16
/ 72
Strong
|
#50
/ 72
Usable
|
DeepSeek V4 Pro |
| Data & Analytics |
#60
/ 110
Excellent
|
#27
/ 110
Excellent
|
GPT-5 Mini |
| AI Strategy |
#68
/ 126
Strong
|
#100
/ 126
Usable
|
DeepSeek V4 Pro |
| Legal & HR |
#40
/ 110
Excellent
|
#10
/ 110
Excellent
|
GPT-5 Mini |
| Summarization & Meeting Notes |
#55
/ 110
Excellent
|
#81
/ 110
Excellent
|
DeepSeek V4 Pro |
| Research & Competitive Analysis |
#44
/ 110
Strong
|
#21
/ 110
Excellent
|
GPT-5 Mini |
| Creative & Comedy | #50 / 110 | #70 / 110 | DeepSeek V4 Pro |
| Translation & Localization |
#32
/ 110
Excellent
|
#52
/ 110
Excellent
|
DeepSeek V4 Pro |
| Investor & Pitch |
#33
/ 66
Strong
|
#50
/ 66
Usable
|
DeepSeek V4 Pro |
| Knowledge & Docs |
#25
/ 110
Strong
|
#41
/ 110
Strong
|
DeepSeek V4 Pro |
| Chef / Home Cooking |
#32
/ 126
Usable
|
#45
/ 126
Usable
|
DeepSeek V4 Pro |
| Product & Project Management |
#12
/ 110
Excellent
|
#1
/ 110
Excellent
|
GPT-5 Mini |
| Presentations & Decks |
#31
/ 110
Excellent
|
#39
/ 110
Excellent
|
DeepSeek V4 Pro |
| Coding |
#43
/ 115
Strong
|
#37
/ 115
Strong
|
GPT-5 Mini |
| Frontend & Landing Pages |
#6
/ 109
Usable
|
#12
/ 109
Needs editing
|
DeepSeek V4 Pro |
| Structured Output |
#23
/ 113
Excellent
|
#17
/ 113
Excellent
|
GPT-5 Mini |
Rank = position among every model config we tested in that task area (lower is better). Sorted by biggest gap first.
Same task, both models — judged
Both models answered the same test case; an independent judge graded each. Quotes are the judge's actual rationale.
Customer Support
Password reset help (Basic Support Reply Test)
“The response perfectly executes the task, providing clear, actionable steps and a strong security caution without inventing unnecessary details. It is highly effective, concise, and ready to use as a template.”
“The model completely missed the format of a customer support reply, instead generating a highly generic, third-party instructional article. It uses placeholders like 'the service's sign-in page' which makes it unusable as a direct response to a user. It also lacks any empathy or conversational tone and is overly verbose.”
Frequently asked
Is DeepSeek V4 Pro better than GPT-5 Mini?
Across 22 task areas we benchmarked, DeepSeek V4 Pro ranks higher in 16 and GPT-5 Mini in 6.
Which is cheaper, DeepSeek V4 Pro or GPT-5 Mini?
DeepSeek V4 Pro costs 1.7× less per token ($1.305 vs $2.25 per 1M).
What is DeepSeek V4 Pro better at?
DeepSeek V4 Pro out-ranks GPT-5 Mini at Executive Assistant, Sales, Customer Support.
What is GPT-5 Mini better at?
GPT-5 Mini out-ranks DeepSeek V4 Pro at Data & Analytics, Legal & HR, Research & Competitive Analysis.
More comparisons
This page is Spring Prompt, running in public
We just did this for every model. Do it for your prompt.
The rankings above come from running real tasks through real models and scoring every output. Spring Prompt is that same engine — pointed at your prompt, your test cases, and your definition of good.
- Generate test cases from your prompt — no eval set required to start.
- Compare models side by side with quality, cost and latency in one matrix.
- Optimise the winner until the scores say it's ready to ship.
Prompt × model results
12 test cases · 3 evals