Meta Llama 3 70B Instruct vs Mixtral 8x22B Instruct v0.1
Meta Llama 3 70B Instruct leads llmrun Score 74.3 to 71.6 (ranges overlap); neither fits 24 GB.
Model B · Mistral AI
Mixtral 8x22B Instruct v0.1140.6B · 39.2B active · 64K context · Apache 2.0 · Apr 2024
Choose Meta Llama 3 70B Instruct if you want…
Higher llmrun Score
74.3 vs 71.6, ranges overlap
Fits a 48 GB GPU
Q4_K_M · 51.3 GB; Mixtral 8x22B Instruct v0.1 needs 42.7 GB at its smallest quant
Fits a Mac 64 GB
Q4_K_M · 51.3 GB; Mixtral 8x22B Instruct v0.1 needs 42.7 GB at its smallest quant
Lighter on memory
46.6 vs 85.2 GB at Q4_K_M
Choose Mixtral 8x22B Instruct v0.1 if you want…
Also does function-calling
Meta Llama 3 70B Instruct doesn't list it
More permissive license
Apache 2.0 vs Llama 3 Community
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
Meta Llama 3 70B Instruct
Doesn't fit
smallest quant needs 28 GB
Mixtral 8x22B Instruct v0.1
Doesn't fit
smallest quant needs 42.7 GB
12 GB
RTX 3060 12GB · 12 GB tier
Meta Llama 3 70B Instruct
Doesn't fit
smallest quant needs 28 GB
Mixtral 8x22B Instruct v0.1
Doesn't fit
smallest quant needs 42.7 GB
16 GB
RTX 5060 Ti 16GB · 16 GB tier
Meta Llama 3 70B Instruct
Doesn't fit
smallest quant needs 28 GB
Mixtral 8x22B Instruct v0.1
Doesn't fit
smallest quant needs 42.7 GB
24 GBreference
RTX 4090 · 24 GB tier
Meta Llama 3 70B Instruct
Doesn't fit
smallest quant needs 28 GB
Mixtral 8x22B Instruct v0.1
Doesn't fit
smallest quant needs 42.7 GB
32 GB
RTX 5090 · 32 GB tier
Meta Llama 3 70B Instruct
IQ2_XS · ~50 tok/s
28 GB · up to 16Kfits
Mixtral 8x22B Instruct v0.1
Doesn't fit
smallest quant needs 42.7 GB
48 GB
RTX A6000 · 48 GB tier
Meta Llama 3 70B Instruct
Q4_K_M · ~11 tok/stight
51.3 GB · up to 4K
Mixtral 8x22B Instruct v0.1
IQ2_XS · ~37 tok/s
46.3 GB · up to 16Kmore headroom
Mac 32 GB
Mac Mini M4 · Macs
Meta Llama 3 70B Instruct
IQ2_XS · ~4 tok/s
28 GB · up to 16Kfits
Mixtral 8x22B Instruct v0.1
Doesn't fit
smallest quant needs 42.7 GB
Mac 64 GB
Mac Studio M4 Max · Macs
Meta Llama 3 70B Instruct
Q4_K_M · ~8 tok/s
51.3 GB · up to 32Kmore headroom
Mixtral 8x22B Instruct v0.1
IQ2_M · ~22 tok/s
51.5 GB · up to 32K
Mac 128 GB
Mac Studio M4 Max · Macs
Meta Llama 3 70B Instruct
Q4_K_M · ~8 tok/s
51.3 GB · up to 32K
Mixtral 8x22B Instruct v0.1
Q4_K_M · ~14 tok/s
88.4 GB · up to 64Kfaster
Reference GPU: RTX 4090. Tinted cell = the better pick on that machine (fits, more headroom, or at least 10% faster).
Overall
Meta Llama 3 70B Instruct
Mixtral 8x22B Instruct v0.1
Reasoning
Meta Llama 3 70B Instruct
no data
Mixtral 8x22B Instruct v0.1
Meta Llama 3 70B Instruct · #89 of 100 open models · #39 of 49 that fit ≤64 GB
Mixtral 8x22B Instruct v0.1 · #92 of 100 open models · #53 of 60 that fit ≤96 GB
Benchmarks
4 shared boards: Meta Llama 3 70B Instruct leads 1, Mixtral 8x22B Instruct v0.1 leads 2 · Benchmarks imported Oct 2026
Meta Llama 3 70B Instruct
40.6%#70/98 open
IndependentEpoch AI · Apr 2024
Mixtral 8x22B Instruct v0.1
34.1%#76/98 open
IndependentEpoch AI · Apr 2024
Meta Llama 3 70B Instruct
56.2%#64/145 open
IndependentTIGER-Lab
Mixtral 8x22B Instruct v0.1
56.3%#63/145 open
IndependentTIGER-Lab
Meta Llama 3 70B Instruct
22.6%#30/39 open
IndependentEpoch AI · Apr 2024
Mixtral 8x22B Instruct v0.1
24.2%#27/39 open
IndependentEpoch AI · Apr 2024
Meta Llama 3 70B Instruct
54.2%#56/68 open
Lab-reportedEpoch AI · Apr 2024
Mixtral 8x22B Instruct v0.1
55.1%#54/68 open
Lab-reportedEpoch AI · Apr 2024
5 more boards (other shared and single-model boards)
Meta Llama 3 70B Instruct
1276.6#119/187 open
ArenaLMArena · Oct 2026
Mixtral 8x22B Instruct v0.1
1229.7#135/187 open
ArenaLMArena · Oct 2026
Only Meta Llama 3 70B Instruct (2)
- AIME 2024/20254.3%
- MMLU79.3%
Only Mixtral 8x22B Instruct v0.1 (2)
- WeirdML3.2%
- ForecastBench56.3
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
Meta Llama 3 70B Instruct
33.0 GB26.4 GB file
32K 43.1 GB · 128K 75.5 GB
Mixtral 8x22B Instruct v0.1
60.5 GB52.1 GB file
32K 67.6 GB · 128K —
Q3_K_S
Meta Llama 3 70B Instruct
34.0 GB30.9 GB file
32K 44.1 GB · 128K 76.4 GB
Mixtral 8x22B Instruct v0.1
62.3 GB61.5 GB file
32K 69.4 GB · 128K —
Q3_K_M
Meta Llama 3 70B Instruct
37.8 GB34.3 GB file
32K 48.0 GB · 128K 80.3 GB
Mixtral 8x22B Instruct v0.1
69.3 GB67.8 GB file
32K 76.4 GB · 128K —
Q4_0
Meta Llama 3 70B Instruct
38.8 GB40.0 GB file
32K 48.9 GB · 128K 81.3 GB
Mixtral 8x22B Instruct v0.1
not listed
Q4_K_M
Meta Llama 3 70B Instruct
46.6 GB42.5 GB file
32K 56.7 GB · 128K 89.1 GB
Mixtral 8x22B Instruct v0.1
85.2 GB85.6 GB file
32K 92.2 GB · 128K —
Q5_K_M
Meta Llama 3 70B Instruct
55.3 GB50.0 GB file
32K 65.4 GB · 128K 97.8 GB
Mixtral 8x22B Instruct v0.1
101 GB100.0 GB file
32K 108 GB · 128K —
Q6_K
Meta Llama 3 70B Instruct
64.0 GB57.9 GB file
32K 74.2 GB · 128K 107 GB
Mixtral 8x22B Instruct v0.1
117 GB~116.0 GB file
32K 124 GB · 128K —
Q8_0
Meta Llama 3 70B Instruct
77.6 GB75.0 GB file
32K 87.7 GB · 128K 120 GB
Mixtral 8x22B Instruct v0.1
141 GB149.4 GB file
32K 148 GB · 128K —
13 more quants
IQ2_XXS
Meta Llama 3 70B Instruct
not listed
Mixtral 8x22B Instruct v0.1
39.4 GB37.9 GB file
32K 46.5 GB · 128K —
IQ2_XS
Meta Llama 3 70B Instruct
23.3 GB21.1 GB file
32K 33.4 GB · 128K 65.8 GB
Mixtral 8x22B Instruct v0.1
43.0 GB42.0 GB file
32K 50.0 GB · 128K —
IQ2_S
Meta Llama 3 70B Instruct
not listed
Mixtral 8x22B Instruct v0.1
44.7 GB42.6 GB file
32K 51.8 GB · 128K —
IQ2_M
Meta Llama 3 70B Instruct
not listed
Mixtral 8x22B Instruct v0.1
48.2 GB46.7 GB file
32K 55.3 GB · 128K —
IQ3_XS
Meta Llama 3 70B Instruct
not listed
Mixtral 8x22B Instruct v0.1
58.8 GB58.2 GB file
32K 65.8 GB · 128K —
Q3_K_L
Meta Llama 3 70B Instruct
39.8 GB37.1 GB file
32K 49.9 GB · 128K 82.3 GB
Mixtral 8x22B Instruct v0.1
72.8 GB72.6 GB file
32K 79.9 GB · 128K —
IQ4_XS
Meta Llama 3 70B Instruct
not listed
Mixtral 8x22B Instruct v0.1
76.4 GB76.4 GB file
32K 83.4 GB · 128K —
Q4_1
Meta Llama 3 70B Instruct
43.7 GB44.3 GB file
32K 53.8 GB · 128K 86.2 GB
Mixtral 8x22B Instruct v0.1
not listed
Q4_K_S
Meta Llama 3 70B Instruct
43.7 GB40.3 GB file
32K 53.8 GB · 128K 86.2 GB
Mixtral 8x22B Instruct v0.1
79.9 GB80.5 GB file
32K 86.9 GB · 128K —
Q5_0
Meta Llama 3 70B Instruct
48.5 GB48.7 GB file
32K 58.6 GB · 128K 91.0 GB
Mixtral 8x22B Instruct v0.1
not listed
Q5_1
Meta Llama 3 70B Instruct
53.4 GB53.0 GB file
32K 63.5 GB · 128K 95.9 GB
Mixtral 8x22B Instruct v0.1
not listed
Q5_K_S
Meta Llama 3 70B Instruct
53.4 GB48.7 GB file
32K 63.5 GB · 128K 95.9 GB
Mixtral 8x22B Instruct v0.1
97.5 GB97.0 GB file
32K 105 GB · 128K —
BF16
Meta Llama 3 70B Instruct
155 GB~141.1 GB file
32K 165 GB · 128K 198 GB
Mixtral 8x22B Instruct v0.1
282 GB~281.3 GB file
32K 289 GB · 128K —
Meta Llama 3 70B Instruct
Ollama · Q4_K_M
ollama run hf.co/second-state/Meta-Llama-3-70B-Instruct-GGUF:Q4_K_MMixtral 8x22B Instruct v0.1
Ollama · Q4_K_M
ollama run hf.co/MaziyarPanahi/Mixtral-8x22B-Instruct-v0.1-GGUF:Q4_K_M| Spec | Meta Llama 3 70B Instruct | Mixtral 8x22B Instruct v0.1 |
|---|---|---|
| Publisher | Meta | Mistral AI |
| Parameters | 70.6B | 140.6B |
| Active parameters | — | 39B per token |
| Architecture | — | MixtralForCausalLM |
| Context | — | 64K tokens |
| Modalities | Chat | Chat, Function-calling |
| License | Llama 3 Community | Apache 2.0 |
| Released | Apr 2024 | Apr 2024 |
Frequently Asked Questions
Which is better, Meta Llama 3 70B Instruct or Mixtral 8x22B Instruct v0.1?
Meta Llama 3 70B Instruct leads llmrun Score 74.3 to 71.6 (ranges overlap); neither fits 24 GB. Choose Meta Llama 3 70B Instruct for higher llmrun Score and fits a 48 GB GPU; choose Mixtral 8x22B Instruct v0.1 for also does function-calling and more permissive license.
Which needs less VRAM, Meta Llama 3 70B Instruct or Mixtral 8x22B Instruct v0.1?
At Q4_K_M, Meta Llama 3 70B Instruct requires 46.6 GB while Mixtral 8x22B Instruct v0.1 needs 85.2 GB, making Meta Llama 3 70B Instruct the easier model to run on your local setup.
Can I run Meta Llama 3 70B Instruct or Mixtral 8x22B Instruct v0.1 on a 16 GB GPU?
Meta Llama 3 70B Instruct does not fit (smallest quant needs 28 GB); Mixtral 8x22B Instruct v0.1 does not fit (smallest quant needs 42.7 GB).