DeepSeek V3.1 vs Kimi K2 Instruct 0905
Kimi K2 Instruct 0905 leads llmrun Score 102.4 to 99.8 (ranges overlap); neither fits 24 GB.
Model A · DeepSeek
DeepSeek V3.1684.5B · 40.1B active · 160K context · MIT · Aug 2025
Model B · Moonshot AI
Kimi K2 Instruct 09051026.5B · 32.9B active · 256K context · Other · Sep 2025
Choose DeepSeek V3.1 if you want…
Lighter on memory
414.6 vs 619.8 GB at Q4_K_M
More permissive license
MIT vs Other
Choose Kimi K2 Instruct 0905 if you want…
Higher llmrun Score
102.4 vs 99.8, ranges overlap
Longer context window
256K vs 160K tokens
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 V3.1
Doesn't fit
smallest quant needs 189.7 GB
Kimi K2 Instruct 0905
Doesn't fit
smallest quant needs 283.7 GB
12 GB
RTX 3060 12GB · 12 GB tier
DeepSeek V3.1
Doesn't fit
smallest quant needs 189.7 GB
Kimi K2 Instruct 0905
Doesn't fit
smallest quant needs 283.7 GB
16 GB
RTX 5060 Ti 16GB · 16 GB tier
DeepSeek V3.1
Doesn't fit
smallest quant needs 189.7 GB
Kimi K2 Instruct 0905
Doesn't fit
smallest quant needs 283.7 GB
24 GBreference
RTX 4090 · 24 GB tier
DeepSeek V3.1
Doesn't fit
smallest quant needs 189.7 GB
Kimi K2 Instruct 0905
Doesn't fit
smallest quant needs 283.7 GB
32 GB
RTX 5090 · 32 GB tier
DeepSeek V3.1
Doesn't fit
smallest quant needs 189.7 GB
Kimi K2 Instruct 0905
Doesn't fit
smallest quant needs 283.7 GB
48 GB
RTX A6000 · 48 GB tier
DeepSeek V3.1
Doesn't fit
smallest quant needs 189.7 GB
Kimi K2 Instruct 0905
Doesn't fit
smallest quant needs 283.7 GB
Mac 32 GB
Mac Mini M4 · Macs
DeepSeek V3.1
Doesn't fit
smallest quant needs 189.7 GB
Kimi K2 Instruct 0905
Doesn't fit
smallest quant needs 283.7 GB
Mac 64 GB
Mac Studio M4 Max · Macs
DeepSeek V3.1
Doesn't fit
smallest quant needs 189.7 GB
Kimi K2 Instruct 0905
Doesn't fit
smallest quant needs 283.7 GB
Mac 128 GB
Mac Studio M4 Max · Macs
DeepSeek V3.1
Doesn't fit
smallest quant needs 189.7 GB
Kimi K2 Instruct 0905
Doesn't fit
smallest quant needs 283.7 GB
Reference GPU: RTX 4090. Tinted cell = the better pick on that machine (fits, more headroom, or at least 10% faster).
Overall
DeepSeek V3.1
Kimi K2 Instruct 0905
Coding
DeepSeek V3.1
no data
Kimi K2 Instruct 0905
Reasoning
DeepSeek V3.1
Kimi K2 Instruct 0905
no data
DeepSeek V3.1 · #47 of 92 open models · #44 of 88 that fit ≤512 GB
Kimi K2 Instruct 0905 · #43 of 92 open models · #40 of 88 that fit ≤640 GB
Benchmarks
2 shared boards: DeepSeek V3.1 leads 1, Kimi K2 Instruct 0905 leads 0 · Benchmarks imported Oct 2026
DeepSeek V3.1
1417.2#36/186 open
ArenaLMArena · Oct 2026
Kimi K2 Instruct 0905
1418.5#34/186 open
ArenaLMArena · Oct 2026
DeepSeek V3.1
38.4%#27/46 open
IndependentEpoch AI · Aug 2025
Kimi K2 Instruct 0905
36.7%#31/46 open
IndependentEpoch AI · Sep 2025
9 more boards (other shared and single-model boards)
DeepSeek V3.1
52.8%#17/25 open
IndependentEpoch AI · Aug 2025
Kimi K2 Instruct 0905
66.7%#10/25 open
IndependentEpoch AI · Sep 2025
DeepSeek V3.1
58.0#14/26 open
Lab-reportedEpoch AI · Aug 2025
Kimi K2 Instruct 0905
59.8#7/26 open
Lab-reportedEpoch AI · Sep 2025
Only DeepSeek V3.1 (4)
- MMLU-Pro84.8%
- SimpleBench40.0%
- DTBench82.7%
- LMCA24.3%
Only Kimi K2 Instruct 0905 (3)
- SWE-bench Verified71.2%
- Aider Polyglot59.1%
- ALE-Bench267.1
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 V3.1
295 GB245.7 GB file
32K 294 GB · 128K 300 GB
Kimi K2 Instruct 0905
440 GB374.1 GB file
32K 439 GB · 128K 446 GB
Q3_K_S
DeepSeek V3.1
303 GB290.4 GB file
32K 302 GB · 128K 309 GB
Kimi K2 Instruct 0905
453 GB443.2 GB file
32K 452 GB · 128K 459 GB
Q3_K_M
DeepSeek V3.1
338 GB320.5 GB file
32K 336 GB · 128K 343 GB
Kimi K2 Instruct 0905
504 GB490.0 GB file
32K 503 GB · 128K 510 GB
Q4_0
DeepSeek V3.1
346 GB380.1 GB file
32K 345 GB · 128K 352 GB
Kimi K2 Instruct 0905
517 GB581.2 GB file
32K 516 GB · 128K 523 GB
Q4_K_M
DeepSeek V3.1
415 GB405.4 GB file
32K 413 GB · 128K 420 GB
Kimi K2 Instruct 0905
620 GB621.2 GB file
32K 618 GB · 128K 625 GB
Q5_K_M
DeepSeek V3.1
492 GB476.2 GB file
32K 490 GB · 128K 497 GB
Kimi K2 Instruct 0905
735 GB728.7 GB file
32K 734 GB · 128K 741 GB
Q6_K
DeepSeek V3.1
569 GB551.3 GB file
32K 567 GB · 128K 574 GB
Kimi K2 Instruct 0905
851 GB842.9 GB file
32K 849 GB · 128K 856 GB
Q8_0
DeepSeek V3.1
688 GB713.3 GB file
32K 687 GB · 128K 694 GB
Kimi K2 Instruct 0905
1030 GB1091.1 GB file
32K 1029 GB · 128K 1036 GB
14 more quants
IQ2_XXS
DeepSeek V3.1
192 GB225.9 GB file
32K 191 GB · 128K 198 GB
Kimi K2 Instruct 0905
286 GB330.8 GB file
32K 285 GB · 128K 292 GB
IQ2_XS
DeepSeek V3.1
209 GB188.4 GB file
32K 208 GB · 128K 215 GB
Kimi K2 Instruct 0905
312 GB283.4 GB file
32K 311 GB · 128K 317 GB
IQ2_S
DeepSeek V3.1
218 GB189.6 GB file
32K 217 GB · 128K 223 GB
Kimi K2 Instruct 0905
325 GB284.1 GB file
32K 323 GB · 128K 330 GB
IQ2_M
DeepSeek V3.1
235 GB235.5 GB file
32K 234 GB · 128K 241 GB
Kimi K2 Instruct 0905
350 GB349.1 GB file
32K 349 GB · 128K 356 GB
IQ3_XXS
DeepSeek V3.1
269 GB279.7 GB file
32K 268 GB · 128K 275 GB
Kimi K2 Instruct 0905
402 GB417.3 GB file
32K 400 GB · 128K 407 GB
IQ3_XS
DeepSeek V3.1
286 GB277.1 GB file
32K 285 GB · 128K 292 GB
Kimi K2 Instruct 0905
427 GB420.7 GB file
32K 426 GB · 128K 433 GB
IQ3_M
DeepSeek V3.1
312 GB307.9 GB file
32K 311 GB · 128K 318 GB
Kimi K2 Instruct 0905
466 GB467.5 GB file
32K 465 GB · 128K 471 GB
Q3_K_L
DeepSeek V3.1
355 GB319.7 GB file
32K 353 GB · 128K 360 GB
Kimi K2 Instruct 0905
530 GB487.7 GB file
32K 529 GB · 128K 536 GB
IQ4_XS
DeepSeek V3.1
372 GB358.3 GB file
32K 371 GB · 128K 377 GB
Kimi K2 Instruct 0905
556 GB547.2 GB file
32K 554 GB · 128K 561 GB
IQ4_NL
DeepSeek V3.1
389 GB379.1 GB file
32K 388 GB · 128K 395 GB
Kimi K2 Instruct 0905
581 GB579.1 GB file
32K 580 GB · 128K 587 GB
Q4_1
DeepSeek V3.1
389 GB420.8 GB file
32K 388 GB · 128K 395 GB
Kimi K2 Instruct 0905
581 GB642.9 GB file
32K 580 GB · 128K 587 GB
Q4_K_S
DeepSeek V3.1
389 GB381.1 GB file
32K 388 GB · 128K 395 GB
Kimi K2 Instruct 0905
581 GB583.3 GB file
32K 580 GB · 128K 587 GB
Q5_K_S
DeepSeek V3.1
475 GB462.6 GB file
32K 473 GB · 128K 480 GB
Kimi K2 Instruct 0905
710 GB707.0 GB file
32K 708 GB · 128K 715 GB
BF16
DeepSeek V3.1
1373 GB1342.3 GB file
32K 1372 GB · 128K 1379 GB
Kimi K2 Instruct 0905
2057 GB2053.2 GB file
32K 2056 GB · 128K 2062 GB
DeepSeek V3.1
Ollama · Q4_K_M
ollama run hf.co/unsloth/DeepSeek-V3.1-GGUF:Q4_K_MKimi K2 Instruct 0905
Ollama · Q4_K_M
ollama run hf.co/unsloth/Kimi-K2-Instruct-0905-GGUF:Q4_K_M| Spec | DeepSeek V3.1 | Kimi K2 Instruct 0905 |
|---|---|---|
| Publisher | DeepSeek | Moonshot AI |
| Parameters | 684.5B | 1026.5B |
| Active parameters | 40B per token | 33B per token |
| Architecture | DeepseekV3ForCausalLM | DeepseekV3ForCausalLM |
| Context | 160K tokens | 256K tokens |
| Modalities | Chat, Function-calling | Chat, Function-calling |
| License | MIT | Other |
| Released | Aug 2025 | Sep 2025 |
Frequently Asked Questions
Which is better, DeepSeek V3.1 or Kimi K2 Instruct 0905?
Kimi K2 Instruct 0905 leads llmrun Score 102.4 to 99.8 (ranges overlap); neither fits 24 GB. Choose DeepSeek V3.1 for lighter on memory and more permissive license; choose Kimi K2 Instruct 0905 for higher llmrun Score and longer context window.
Which needs less VRAM, DeepSeek V3.1 or Kimi K2 Instruct 0905?
At Q4_K_M, DeepSeek V3.1 requires 414.6 GB while Kimi K2 Instruct 0905 needs 619.8 GB, making DeepSeek V3.1 the easier model to run on your local setup.
Can I run DeepSeek V3.1 or Kimi K2 Instruct 0905 on a 16 GB GPU?
DeepSeek V3.1 does not fit (smallest quant needs 189.7 GB); Kimi K2 Instruct 0905 does not fit (smallest quant needs 283.7 GB).
Which has a longer context window, DeepSeek V3.1 or Kimi K2 Instruct 0905?
DeepSeek V3.1 supports 163,840 tokens and Kimi K2 Instruct 0905 supports 262,144 tokens.