DeepSeek V3.2
VS
MiniMax M3
DeepSeek V3.2 vs MiniMax M3: 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.
MiniMax M3 wins 18 of 22 task areas we tested; DeepSeek V3.2 takes 4. DeepSeek V3.2 costs 2.6× less per token ($0.572 vs $1.5 per 1M).
DeepSeek V3.2 costs 2.6× less per token ($0.572 vs $1.5 per 1M).
Task by task
| Task area | DeepSeek V3.2 | MiniMax M3 | Winner |
|---|---|---|---|
| Sales |
#91
/ 110
Usable
|
#2
/ 110
Strong
|
MiniMax M3 |
| AI Strategy |
#102
/ 126
Usable
|
#27
/ 126
Strong
|
MiniMax M3 |
| Data & Analytics |
#69
/ 110
Excellent
|
#1
/ 110
Excellent
|
MiniMax M3 |
| Content & Brand |
#95
/ 124
Usable
|
#28
/ 124
Strong
|
MiniMax M3 |
| Legal & HR |
#90
/ 110
Strong
|
#34
/ 110
Excellent
|
MiniMax M3 |
| Presentations & Decks |
#81
/ 110
Strong
|
#29
/ 110
Excellent
|
MiniMax M3 |
| Coding |
#100
/ 115
Usable
|
#49
/ 115
Strong
|
MiniMax M3 |
| Summarization & Meeting Notes |
#30
/ 110
Excellent
|
#79
/ 110
Excellent
|
DeepSeek V3.2 |
| Landing Pages |
#48
/ 72
Usable
|
#2
/ 72
Strong
|
MiniMax M3 |
| Executive Assistant |
#76
/ 112
Usable
|
#39
/ 112
Strong
|
MiniMax M3 |
| Investor & Pitch |
#46
/ 66
Usable
|
#12
/ 66
Strong
|
MiniMax M3 |
| Translation & Localization |
#89
/ 110
Strong
|
#55
/ 110
Excellent
|
MiniMax M3 |
| Creative & Comedy | #87 / 110 | #55 / 110 | MiniMax M3 |
| Research & Competitive Analysis |
#53
/ 110
Usable
|
#33
/ 110
Strong
|
MiniMax M3 |
| Product & Project Management |
#21
/ 110
Excellent
|
#3
/ 110
Excellent
|
MiniMax M3 |
| RAG, Safety & Grounding |
#83
/ 113
Excellent
|
#72
/ 113
Excellent
|
MiniMax M3 |
| Customer Support |
#74
/ 113
Usable
|
#83
/ 113
Usable
|
DeepSeek V3.2 |
| Knowledge & Docs |
#58
/ 110
Usable
|
#67
/ 110
Usable
|
DeepSeek V3.2 |
| Frontend & Landing Pages |
#25
/ 109
Needs editing
|
#20
/ 109
Needs editing
|
MiniMax M3 |
| Chef / Home Cooking |
#78
/ 126
Usable
|
#74
/ 126
Usable
|
MiniMax M3 |
| Training & Education |
#38
/ 110
Excellent
|
#42
/ 110
Excellent
|
DeepSeek V3.2 |
| Structured Output |
#67
/ 113
Strong
|
#65
/ 113
Strong
|
MiniMax M3 |
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.
Coding
Unit tests for parser (Code Quality and Testing Test)
“The response contains multiple broken tests that will fail when run against the provided parser. It hallucinates random Chinese text into a JSON string, misunderstands how Python's `dict.get()` handles explicit nulls, misunderstands `str(None)`, and fails to provide the requested brief explanation of reasoning.”
“The response is expert-level and production-ready. It fulfills all requirements perfectly, uses advanced pytest features appropriately, and provides insightful commentary on the function's limitations without overcomplicating the requested tests.”
Frequently asked
Is DeepSeek V3.2 better than MiniMax M3?
Across 22 task areas we benchmarked, MiniMax M3 ranks higher in 18 and DeepSeek V3.2 in 4.
Which is cheaper, DeepSeek V3.2 or MiniMax M3?
DeepSeek V3.2 costs 2.6× less per token ($0.572 vs $1.5 per 1M).
What is DeepSeek V3.2 better at?
DeepSeek V3.2 out-ranks MiniMax M3 at Summarization & Meeting Notes, Customer Support, Knowledge & Docs.
What is MiniMax M3 better at?
MiniMax M3 out-ranks DeepSeek V3.2 at Sales, AI Strategy, Data & Analytics.
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