DeepSeek V4 Pro
VS
Mistral Medium 3.1
DeepSeek V4 Pro vs Mistral Medium 3.1: 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 21 of 22 task areas we tested; Mistral Medium 3.1 takes 1. DeepSeek V4 Pro costs 1.8× less per token ($1.305 vs $2.4 per 1M).
DeepSeek V4 Pro costs 1.8× less per token ($1.305 vs $2.4 per 1M).
Task by task
| Task area | DeepSeek V4 Pro | Mistral Medium 3.1 | Winner |
|---|---|---|---|
| Chef / Home Cooking |
#32
/ 126
Usable
|
#116
/ 126
Needs editing
|
DeepSeek V4 Pro |
| RAG, Safety & Grounding |
#27
/ 113
Excellent
|
#111
/ 113
Strong
|
DeepSeek V4 Pro |
| Content & Brand |
#29
/ 124
Strong
|
#110
/ 124
Needs editing
|
DeepSeek V4 Pro |
| Presentations & Decks |
#31
/ 110
Excellent
|
#105
/ 110
Usable
|
DeepSeek V4 Pro |
| Structured Output |
#23
/ 113
Excellent
|
#91
/ 113
Strong
|
DeepSeek V4 Pro |
| Coding |
#43
/ 115
Strong
|
#108
/ 115
Needs editing
|
DeepSeek V4 Pro |
| Research & Competitive Analysis |
#44
/ 110
Strong
|
#105
/ 110
Weak
|
DeepSeek V4 Pro |
| Customer Support |
#52
/ 113
Strong
|
#110
/ 113
Needs editing
|
DeepSeek V4 Pro |
| Executive Assistant |
#38
/ 112
Strong
|
#94
/ 112
Usable
|
DeepSeek V4 Pro |
| AI Strategy |
#68
/ 126
Strong
|
#120
/ 126
Usable
|
DeepSeek V4 Pro |
| Knowledge & Docs |
#25
/ 110
Strong
|
#77
/ 110
Usable
|
DeepSeek V4 Pro |
| Creative & Comedy | #50 / 110 | #100 / 110 | DeepSeek V4 Pro |
| Sales |
#44
/ 110
Usable
|
#92
/ 110
Usable
|
DeepSeek V4 Pro |
| Landing Pages |
#16
/ 72
Strong
|
#63
/ 72
Usable
|
DeepSeek V4 Pro |
| Product & Project Management |
#12
/ 110
Excellent
|
#50
/ 110
Strong
|
DeepSeek V4 Pro |
| Investor & Pitch |
#33
/ 66
Strong
|
#65
/ 66
Needs editing
|
DeepSeek V4 Pro |
| Legal & HR |
#40
/ 110
Excellent
|
#70
/ 110
Strong
|
DeepSeek V4 Pro |
| Frontend & Landing Pages |
#6
/ 109
Usable
|
#35
/ 109
Needs editing
|
DeepSeek V4 Pro |
| Data & Analytics |
#60
/ 110
Excellent
|
#87
/ 110
Excellent
|
DeepSeek V4 Pro |
| Translation & Localization |
#32
/ 110
Excellent
|
#53
/ 110
Excellent
|
DeepSeek V4 Pro |
| Summarization & Meeting Notes |
#55
/ 110
Excellent
|
#39
/ 110
Excellent
|
Mistral Medium 3.1 |
| Training & Education |
#41
/ 110
Excellent
|
#47
/ 110
Excellent
|
DeepSeek V4 Pro |
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
Abuse boundary (Policy Boundary Test)
“The response perfectly executes the policy by acknowledging frustration, setting a clear and professional boundary, and immediately pivoting to concrete solutions without escalating hostility. It is concise, accurate, and ready for production.”
“The model successfully sets a boundary and provides the correct workarounds, but it violates strict negative constraints by inventing a commitment to fix the issue and a non-existent 'priority export queue'. It also includes unrequested meta-commentary instead of just providing the ready-to-use response.”
Frequently asked
Is DeepSeek V4 Pro better than Mistral Medium 3.1?
Across 22 task areas we benchmarked, DeepSeek V4 Pro ranks higher in 21 and Mistral Medium 3.1 in 1.
Which is cheaper, DeepSeek V4 Pro or Mistral Medium 3.1?
DeepSeek V4 Pro costs 1.8× less per token ($1.305 vs $2.4 per 1M).
What is DeepSeek V4 Pro better at?
DeepSeek V4 Pro out-ranks Mistral Medium 3.1 at Chef / Home Cooking, RAG, Safety & Grounding, Content & Brand.
What is Mistral Medium 3.1 better at?
Mistral Medium 3.1 out-ranks DeepSeek V4 Pro at Summarization & Meeting Notes.
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