DeepSeek V4 Pro vs Kimi K3
Kimi K3 leads llmrun Score 132.8 to 116.5; neither fits 24 GB.
Model A · DeepSeek
DeepSeek V4 Pro1598.8B · 50.7B active · 1M context · MIT · Apr 2026
Model B · Moonshot AI
Kimi K32779.9B · 105.8B active · 1M context · Other · Jun 2026
Choose DeepSeek V4 Pro if you want…
Lighter on memory
959.6 vs 1673.7 GB at Q4_K_M
More permissive license
MIT vs Other
Choose Kimi K3 if you want…
Higher llmrun Score
132.8 vs 116.5
Stronger at coding
134.4 vs 110.6
Stronger at math
134.7 vs 117.6
Ahead on the benchmarks
20 of 21 shared boards
Also does vision
DeepSeek V4 Pro doesn't list it
On your hardware
Each model at its default quant (Q4_K_M), a smaller one when it must. Estimates, ±20%.
8 GB
RTX 4060 · 8 GB tier
DeepSeek V4 Pro
Doesn't fit
smallest quant needs 680 GB
Kimi K3
Doesn't fit
smallest quant needs 766.5 GB
12 GB
RTX 3060 12GB · 12 GB tier
DeepSeek V4 Pro
Doesn't fit
smallest quant needs 680 GB
Kimi K3
Doesn't fit
smallest quant needs 766.5 GB
16 GB
RTX 5060 Ti 16GB · 16 GB tier
DeepSeek V4 Pro
Doesn't fit
smallest quant needs 680 GB
Kimi K3
Doesn't fit
smallest quant needs 766.5 GB
24 GBreference
RTX 4090 · 24 GB tier
DeepSeek V4 Pro
Doesn't fit
smallest quant needs 680 GB
Kimi K3
Doesn't fit
smallest quant needs 766.5 GB
32 GB
RTX 5090 · 32 GB tier
DeepSeek V4 Pro
Doesn't fit
smallest quant needs 680 GB
Kimi K3
Doesn't fit
smallest quant needs 766.5 GB
48 GB
RTX A6000 · 48 GB tier
DeepSeek V4 Pro
Doesn't fit
smallest quant needs 680 GB
Kimi K3
Doesn't fit
smallest quant needs 766.5 GB
Mac 32 GB
Mac Mini M4 · Macs
DeepSeek V4 Pro
Doesn't fit
smallest quant needs 680 GB
Kimi K3
Doesn't fit
smallest quant needs 766.5 GB
Mac 64 GB
Mac Studio M4 Max · Macs
DeepSeek V4 Pro
Doesn't fit
smallest quant needs 680 GB
Kimi K3
Doesn't fit
smallest quant needs 766.5 GB
Mac 128 GB
Mac Studio M4 Max · Macs
DeepSeek V4 Pro
Doesn't fit
smallest quant needs 680 GB
Kimi K3
Doesn't fit
smallest quant needs 766.5 GB
Reference GPU: RTX 4090. Tinted cell = the better pick on that machine (fits, more headroom, or at least 10% faster).
Overall
DeepSeek V4 Pro
Kimi K3
Coding
DeepSeek V4 Pro
Kimi K3
Agents
DeepSeek V4 Pro
no data
Kimi K3
Math
DeepSeek V4 Pro
Kimi K3
Science
DeepSeek V4 Pro
Kimi K3
Reasoning
DeepSeek V4 Pro
Kimi K3
DeepSeek V4 Pro · #16 of 100 open models · #16 of 99 that fit ≤1 TB
Kimi K3 · #2 of 100 open models · #2 of 99 that fit ≤2 TB
Benchmarks
21 shared boards: DeepSeek V4 Pro leads 1, Kimi K3 leads 20 · Benchmarks imported Oct 2026
DeepSeek V4 Pro
90.9%#5/98 open
IndependentEpoch AI · Apr 2026
Kimi K3
93.1%#1/98 open
IndependentEpoch AI · Jul 2026
DeepSeek V4 Pro
70.0#16/19 open
IndependentLiveBench · Jun 2026
Kimi K3
81.5#1/19 open
IndependentLiveBench · Jun 2026
DeepSeek V4 Pro
82.7#10/19 open
IndependentLiveBench · Jun 2026
Kimi K3
90.7#1/19 open
IndependentLiveBench · Jun 2026
DeepSeek V4 Pro
96.7%#5/86 open
IndependentEpoch AI · Apr 2026
Kimi K3
97.2%#4/86 open
IndependentEpoch AI · Jul 2026
DeepSeek V4 Pro
45.3%#12/17 open
IndependentEpoch AI · Apr 2026
Kimi K3
72.2%#1/17 open
IndependentEpoch AI · Jul 2026
DeepSeek V4 Pro
90.7#3/19 open
IndependentLiveBench · Jun 2026
Kimi K3
84.4#13/19 open
IndependentLiveBench · Jun 2026
DeepSeek V4 Pro
50.0%#12/66 open
Lab-reportedEpoch AI · Apr 2026
Kimi K3
59.5%#2/66 open
Lab-reportedEpoch AI · Jul 2026
DeepSeek V4 Pro
47.0%#4/17 open
IndependentEpoch AI · Apr 2026
Kimi K3
50.6%#3/17 open
IndependentEpoch AI · Jul 2026
DeepSeek V4 Pro
2.4%#13/13 open
IndependentEpoch AI · Apr 2026
Kimi K3
39.0%#1/13 open
IndependentEpoch AI · Jul 2026
DeepSeek V4 Pro
48.9%#10/46 open
IndependentEpoch AI · Apr 2026
Kimi K3
82.6%#1/46 open
IndependentEpoch AI · Jul 2026
28 more boards (other shared and single-model boards)
DeepSeek V4 Pro
17.6%#9/12 open
Lab-reportedEpoch AI · Apr 2026
Kimi K3
44.2%#1/12 open
Lab-reportedEpoch AI · Jul 2026
DeepSeek V4 Pro
1006.1#10/43 open
Lab-reportedEpoch AI · Apr 2026
Kimi K3
1524.5#1/43 open
Lab-reportedEpoch AI · Jul 2026
DeepSeek V4 Pro
19.0%#12/14 open
IndependentEpoch AI · Apr 2026
Kimi K3
37.0%#3/14 open
IndependentEpoch AI · Jul 2026
DeepSeek V4 Pro
12.9%#9/66 open
Lab-reportedEpoch AI · Apr 2026
Kimi K3
23.4%#2/66 open
Lab-reportedEpoch AI · Jul 2026
DeepSeek V4 Pro
90.7%#6/68 open
Lab-reportedEpoch AI · Apr 2026
Kimi K3
91.2%#3/68 open
Lab-reportedEpoch AI · Jul 2026
DeepSeek V4 Pro
55.6#24/26 open
Lab-reportedEpoch AI · Apr 2026
Kimi K3
61.9#1/26 open
Lab-reportedEpoch AI · Jul 2026
DeepSeek V4 Pro
41.2%#8/54 open
Lab-reportedEpoch AI · Apr 2026
Kimi K3
52.7%#2/54 open
Lab-reportedEpoch AI · Jul 2026
DeepSeek V4 Pro
17.0%#11/18 open
IndependentEpoch AI · Apr 2026
Kimi K3
26.0%#5/18 open
IndependentEpoch AI · Jul 2026
DeepSeek V4 Pro
16.0%#14/26 open
Lab-reportedEpoch AI · Apr 2026
Kimi K3
87.0%#1/26 open
Lab-reportedEpoch AI · Jul 2026
DeepSeek V4 Pro
40.0%#9/13 open
Lab-reportedEpoch AI · Apr 2026
Kimi K3
95.0%#1/13 open
Lab-reportedEpoch AI · Jul 2026
DeepSeek V4 Pro
3284.5#8/17 open
Lab-reportedEpoch AI · Apr 2026
Kimi K3
5165.0#5/17 open
Lab-reportedEpoch AI · Jul 2026
DeepSeek V4 Pro
1457.8#11/187 open
ArenaLMArena · Oct 2026
Kimi K3
1487.1#1/187 open
ArenaLMArena · Oct 2026
DeepSeek V4 Pro
1464.2#20/52 open
ArenaLMArena · Oct 2026
Kimi K3
1655.2#1/52 open
ArenaLMArena · Oct 2026
Only DeepSeek V4 Pro (1)
- CL-bench Life13.5%
Only Kimi K3 (14)
- ARC-AGI-260.4%
- ARC-AGI94.5%
- SimpleBench60.7%
- DeepSWE68.5%
- APEX-Agents50.6%
- Furniture Assembly34.2%
- LiveBench 2026 Reasoning90.7
- LiveBench 2026 Browser16.0
- LiveBench 2026 Data Analysis78.7
- FrontierSWE25.9%
- LiveBench 2026 Coding59.3
- LiveBench 2026 Long-Horizon42.8
- LiveBench 2026 Design77.2
- LiveBench 2026 Research79.0
VRAM by quantization
Memory to load the model and 2K context; below it, the total at 32K and 128K context. Lighter side in bold.
Q2_K
DeepSeek V4 Pro
680 GB574.1 GB file
32K 680 GB · 128K 682 GB
Kimi K3
1187 GB928.6 GB file
32K 1185 GB · 128K 1196 GB
Q3_K_M
DeepSeek V4 Pro
780 GB748.7 GB file
32K 780 GB · 128K 782 GB
Kimi K3
1361 GB~1355.2 GB file
32K 1359 GB · 128K 1370 GB
Q4_K_M
DeepSeek V4 Pro
960 GB954.6 GB file
32K 960 GB · 128K 961 GB
Kimi K3
1674 GB~1668.0 GB file
32K 1672 GB · 128K 1682 GB
Q5_K_M
DeepSeek V4 Pro
1140 GB~1139.2 GB file
32K 1140 GB · 128K 1141 GB
Kimi K3
1986 GB~1980.7 GB file
32K 1985 GB · 128K 1995 GB
Q6_K
DeepSeek V4 Pro
1319 GB~1319.0 GB file
32K 1320 GB · 128K 1321 GB
Kimi K3
2299 GB~2293.4 GB file
32K 2297 GB · 128K 2308 GB
Q8_0
DeepSeek V4 Pro
1599 GB1671.8 GB file
32K 1600 GB · 128K 1601 GB
Kimi K3
2786 GB2955.0 GB file
32K 2784 GB · 128K 2794 GB
2 more quants
IQ2_XXS
DeepSeek V4 Pro
not listed
Kimi K3
770 GB711.1 GB file
32K 768 GB · 128K 779 GB
BF16
DeepSeek V4 Pro
3198 GB~3197.7 GB file
32K 3198 GB · 128K 3200 GB
Kimi K3
5566 GB~5559.9 GB file
32K 5564 GB · 128K 5574 GB
DeepSeek V4 Pro
Ollama · Q4_K_M
ollama run hf.co/teamblobfish/DeepSeek-V4-Pro-GGUF:Q4_K_MKimi K3
Ollama · Q4_K_M
ollama run hf.co/unsloth/Kimi-K3-GGUF| Spec | DeepSeek V4 Pro | Kimi K3 |
|---|---|---|
| Publisher | DeepSeek | Moonshot AI |
| Parameters | 1598.8B | 2779.9B |
| Active parameters | 51B per token | 106B per token |
| Architecture | DeepseekV4ForCausalLM | KimiK3ForConditionalGeneration |
| Context | 1M tokens | 1M tokens |
| Modalities | Chat | Vision, Chat |
| License | MIT | Other |
| Released | Apr 2026 | Jun 2026 |
Frequently Asked Questions
Which is better, DeepSeek V4 Pro or Kimi K3?
Kimi K3 leads llmrun Score 132.8 to 116.5; neither fits 24 GB. Choose DeepSeek V4 Pro for lighter on memory and more permissive license; choose Kimi K3 for higher llmrun Score and stronger at coding.
Which needs less VRAM, DeepSeek V4 Pro or Kimi K3?
At Q4_K_M, DeepSeek V4 Pro requires 959.6 GB while Kimi K3 needs 1673.7 GB, making DeepSeek V4 Pro the easier model to run on your local setup.
Can I run DeepSeek V4 Pro or Kimi K3 on a 16 GB GPU?
DeepSeek V4 Pro does not fit (smallest quant needs 680 GB); Kimi K3 does not fit (smallest quant needs 766.5 GB).
Which has a longer context window, DeepSeek V4 Pro or Kimi K3?
DeepSeek V4 Pro supports 1,048,576 tokens and Kimi K3 supports 1,048,576 tokens.