Meta Llama 3 8B Instruct vs GPT OSS 20B
GPT OSS 20B leads llmrun Score 96.4 to 62.0; both fit 24 GB at Q4_K_M; GPT OSS 20B is 1.5× faster on RTX 4090 (~190 vs ~124 tok/s).
Model B · OpenAI
GPT OSS 20B20.9B · 4.2B active · 128K context · Apache 2.0 · Aug 2025
Choose Meta Llama 3 8B Instruct if you want…
Fits an 8 GB GPU
Q4_K_M · 7.9 GB; GPT OSS 20B needs 6.3 GB at its smallest quant
Lighter on memory
5.3 vs 12.9 GB at Q4_K_M
Choose GPT OSS 20B if you want…
Higher llmrun Score
96.4 vs 62.0
Stronger at reasoning
91.3 vs 65.2
Ahead on the benchmarks
4 of 4 shared boards
1.5× faster on RTX 4090
~190 vs ~124 tok/s
Also does function-calling
Meta Llama 3 8B 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 8B Instruct
Q4_K_M · ~33 tok/s
7.9 GB · up to 16Kmore headroom
GPT OSS 20B
IQ2_M · ~103 tok/s
7.6 GB · up to 32K
12 GB
RTX 3060 12GB · 12 GB tier
Meta Llama 3 8B Instruct
Q4_K_M · ~44 tok/s
7.9 GB · up to 32Kmore headroom
GPT OSS 20B
IQ4_XS · ~89 tok/s
11.8 GB · up to 16K
16 GB
RTX 5060 Ti 16GB · 16 GB tier
Meta Llama 3 8B Instruct
Q4_K_M · ~55 tok/s
7.9 GB · up to 32K
GPT OSS 20B
Q4_K_M · ~98 tok/s
13.1 GB · up to 128Kfaster
24 GBreference
RTX 4090 · 24 GB tier
Meta Llama 3 8B Instruct
Q4_K_M · ~124 tok/s
7.9 GB · up to 32K
GPT OSS 20B
Q4_K_M · ~190 tok/s
13.1 GB · up to 128Kfaster
32 GB
RTX 5090 · 32 GB tier
Meta Llama 3 8B Instruct
Q4_K_M · ~220 tok/s
7.9 GB · up to 32K
GPT OSS 20B
Q4_K_M · ~282 tok/s
13.1 GB · up to 128Kfaster
48 GB
RTX A6000 · 48 GB tier
Meta Llama 3 8B Instruct
Q4_K_M · ~94 tok/s
7.9 GB · up to 32K
GPT OSS 20B
Q4_K_M · ~154 tok/s
13.1 GB · up to 128Kfaster
Mac 32 GB
Mac Mini M4 · Macs
Meta Llama 3 8B Instruct
Q4_K_M · ~16 tok/s
7.9 GB · up to 32K
GPT OSS 20B
Q4_K_M · ~28 tok/s
13.1 GB · up to 128Kfaster
Mac 64 GB
Mac Studio M4 Max · Macs
Meta Llama 3 8B Instruct
Q4_K_M · ~72 tok/s
7.9 GB · up to 32K
GPT OSS 20B
Q4_K_M · ~86 tok/s
13.1 GB · up to 128Kfaster
Mac 128 GB
Mac Studio M4 Max · Macs
Meta Llama 3 8B Instruct
Q4_K_M · ~72 tok/s
7.9 GB · up to 32K
GPT OSS 20B
Q4_K_M · ~86 tok/s
13.1 GB · up to 128Kfaster
Reference GPU: RTX 4090. Tinted cell = the better pick on that machine (fits, more headroom, or at least 10% faster).
Overall
Meta Llama 3 8B Instruct
GPT OSS 20B
Coding
Meta Llama 3 8B Instruct
no data
GPT OSS 20B
Science
Meta Llama 3 8B Instruct
no data
GPT OSS 20B
Reasoning
Meta Llama 3 8B Instruct
GPT OSS 20B
Meta Llama 3 8B Instruct · #105 of 106 open models · #36 of 36 that fit 24 GB
GPT OSS 20B · #69 of 106 open models · #16 of 36 that fit 24 GB
Benchmarks
4 shared boards: Meta Llama 3 8B Instruct leads 0, GPT OSS 20B leads 4 · Benchmarks imported Nov 2026
Meta Llama 3 8B Instruct
26.1%#91/100 open
IndependentEpoch AI · Apr 2024
GPT OSS 20B
60.8%#44/100 open
IndependentEpoch AI · Aug 2025
Meta Llama 3 8B Instruct
1.9%#78/88 open
IndependentEpoch AI · Apr 2024
GPT OSS 20B
65.3%#36/88 open
IndependentEpoch AI · Aug 2025
Meta Llama 3 8B Instruct
41.0%#100/146 open
IndependentTIGER-Lab
GPT OSS 20B
73.6%#33/146 open
IndependentTIGER-Lab
Meta Llama 3 8B Instruct
43.9%#68/70 open
Lab-reportedEpoch AI · Apr 2024
GPT OSS 20B
68.0%#36/70 open
Lab-reportedEpoch AI · Aug 2025
14 more boards (other shared and single-model boards)
Meta Llama 3 8B Instruct
1223.7#138/188 open
ArenaLMArena · Oct 2026
GPT OSS 20B
1317.3#98/188 open
ArenaLMArena · Oct 2026
Only Meta Llama 3 8B Instruct (7)
- MATH Level 56.1%
- MMLU68.8%
- ForecastBench59.4
- WinoGrande65.0%
- ARC Challenge82.8%
- OpenBookQA82.6%
- TriviaQA67.7%
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 8B Instruct
3.8 GB3.2 GB file
32K 9.3 GB · 128K 26.9 GB
GPT OSS 20B
9.3 GB11.5 GB file
32K 9.8 GB · 128K 11.5 GB
Q3_K_S
Meta Llama 3 8B Instruct
3.9 GB3.7 GB file
32K 9.4 GB · 128K 27.0 GB
GPT OSS 20B
9.5 GB11.5 GB file
32K 10.0 GB · 128K 11.7 GB
Q3_K_M
Meta Llama 3 8B Instruct
4.3 GB4.0 GB file
32K 9.8 GB · 128K 27.4 GB
GPT OSS 20B
10.6 GB11.5 GB file
32K 11.1 GB · 128K 12.8 GB
Q4_0
Meta Llama 3 8B Instruct
4.4 GB4.7 GB file
32K 9.9 GB · 128K 27.5 GB
GPT OSS 20B
10.8 GB11.5 GB file
32K 11.3 GB · 128K 13.0 GB
Q4_K_M
Meta Llama 3 8B Instruct
5.3 GB4.9 GB file
32K 10.8 GB · 128K 28.4 GB
GPT OSS 20B
12.9 GB11.6 GB file
32K 13.4 GB · 128K 15.1 GB
Q5_K_M
Meta Llama 3 8B Instruct
6.3 GB5.7 GB file
32K 11.8 GB · 128K 29.4 GB
GPT OSS 20B
15.3 GB11.7 GB file
32K 15.8 GB · 128K 17.5 GB
Q6_K
Meta Llama 3 8B Instruct
7.3 GB6.6 GB file
32K 12.8 GB · 128K 30.4 GB
GPT OSS 20B
17.6 GB12.0 GB file
32K 18.1 GB · 128K 19.8 GB
Q8_0
Meta Llama 3 8B Instruct
8.8 GB8.5 GB file
32K 14.3 GB · 128K 32.0 GB
GPT OSS 20B
21.3 GB~20.9 GB file
32K 21.8 GB · 128K 23.5 GB
20 more quants
IQ2_XXS
Meta Llama 3 8B Instruct
2.4 GB2.4 GB file
32K 7.9 GB · 128K 25.6 GB
GPT OSS 20B
6.1 GB~5.8 GB file
32K 6.6 GB · 128K 8.3 GB
IQ2_XS
Meta Llama 3 8B Instruct
2.7 GB2.6 GB file
32K 8.2 GB · 128K 25.8 GB
GPT OSS 20B
6.7 GB~6.3 GB file
32K 7.2 GB · 128K 8.9 GB
IQ2_S
Meta Llama 3 8B Instruct
2.8 GB2.8 GB file
32K 8.3 GB · 128K 25.9 GB
GPT OSS 20B
6.9 GB~6.5 GB file
32K 7.4 GB · 128K 9.1 GB
IQ2_M
Meta Llama 3 8B Instruct
3.0 GB2.9 GB file
32K 8.5 GB · 128K 26.1 GB
GPT OSS 20B
7.4 GB~7.1 GB file
32K 7.9 GB · 128K 9.6 GB
IQ3_XXS
Meta Llama 3 8B Instruct
3.4 GB3.3 GB file
32K 8.9 GB · 128K 26.5 GB
GPT OSS 20B
8.5 GB11.6 GB file
32K 9.0 GB · 128K 10.7 GB
Q2_K_S
Meta Llama 3 8B Instruct
3.5 GB3.0 GB file
32K 9.0 GB · 128K 26.7 GB
GPT OSS 20B
8.7 GB12.1 GB file
32K 9.2 GB · 128K 10.9 GB
IQ3_XS
Meta Llama 3 8B Instruct
3.6 GB3.5 GB file
32K 9.1 GB · 128K 26.8 GB
GPT OSS 20B
9.0 GB11.6 GB file
32K 9.5 GB · 128K 11.2 GB
IQ3_S
Meta Llama 3 8B Instruct
3.8 GB3.7 GB file
32K 9.3 GB · 128K 26.9 GB
GPT OSS 20B
9.3 GB12.1 GB file
32K 9.8 GB · 128K 11.5 GB
IQ3_M
Meta Llama 3 8B Instruct
4.0 GB3.8 GB file
32K 9.5 GB · 128K 27.1 GB
GPT OSS 20B
9.8 GB11.6 GB file
32K 10.3 GB · 128K 12.0 GB
Q3_K_L
Meta Llama 3 8B Instruct
4.5 GB4.3 GB file
32K 10.0 GB · 128K 27.7 GB
GPT OSS 20B
11.1 GB11.5 GB file
32K 11.6 GB · 128K 13.3 GB
IQ4_XS
Meta Llama 3 8B Instruct
4.8 GB4.4 GB file
32K 10.3 GB · 128K 27.9 GB
GPT OSS 20B
11.6 GB11.6 GB file
32K 12.1 GB · 128K 13.8 GB
IQ4_NL
Meta Llama 3 8B Instruct
5.0 GB4.7 GB file
32K 10.5 GB · 128K 28.1 GB
GPT OSS 20B
12.1 GB11.6 GB file
32K 12.6 GB · 128K 14.3 GB
Q4_1
Meta Llama 3 8B Instruct
5.0 GB5.1 GB file
32K 10.5 GB · 128K 28.1 GB
GPT OSS 20B
12.1 GB11.6 GB file
32K 12.6 GB · 128K 14.3 GB
Q4_K_S
Meta Llama 3 8B Instruct
5.0 GB4.7 GB file
32K 10.5 GB · 128K 28.1 GB
GPT OSS 20B
12.1 GB11.6 GB file
32K 12.6 GB · 128K 14.3 GB
Q4_K_L
Meta Llama 3 8B Instruct
not listed
GPT OSS 20B
13.2 GB11.9 GB file
32K 13.7 GB · 128K 15.4 GB
Q5_0
Meta Llama 3 8B Instruct
5.5 GB5.6 GB file
32K 11.0 GB · 128K 28.6 GB
GPT OSS 20B
not listed
Q5_1
Meta Llama 3 8B Instruct
6.1 GB6.1 GB file
32K 11.6 GB · 128K 29.2 GB
GPT OSS 20B
not listed
Q5_K_S
Meta Llama 3 8B Instruct
6.1 GB5.6 GB file
32K 11.6 GB · 128K 29.2 GB
GPT OSS 20B
14.8 GB11.7 GB file
32K 15.3 GB · 128K 17.0 GB
Q5_K_L
Meta Llama 3 8B Instruct
not listed
GPT OSS 20B
15.5 GB11.9 GB file
32K 16.0 GB · 128K 17.7 GB
BF16
Meta Llama 3 8B Instruct
17.7 GB~16.1 GB file
32K 23.2 GB · 128K 40.8 GB
GPT OSS 20B
42.2 GB~41.8 GB file
32K 42.7 GB · 128K 44.4 GB
Meta Llama 3 8B Instruct
Ollama · Q4_K_M
ollama run hf.co/MaziyarPanahi/Meta-Llama-3-8B-Instruct-GGUF:Q4_K_MGPT OSS 20B
Ollama · Q4_K_M
ollama run hf.co/unsloth/gpt-oss-20b-GGUF:Q4_K_M| Spec | Meta Llama 3 8B Instruct | GPT OSS 20B |
|---|---|---|
| Publisher | Meta | OpenAI |
| Parameters | 8.0B | 20.9B |
| Active parameters | — | 4.2B per token |
| Architecture | — | GptOssForCausalLM |
| Context | — | 128K tokens |
| Modalities | Chat | Chat, Function-calling |
| License | Llama 3 Community | Apache 2.0 |
| Released | Apr 2024 | Aug 2025 |
Frequently Asked Questions
Which is better, Meta Llama 3 8B Instruct or GPT OSS 20B?
GPT OSS 20B leads llmrun Score 96.4 to 62.0; both fit 24 GB at Q4_K_M; GPT OSS 20B is 1.5× faster on RTX 4090 (~190 vs ~124 tok/s). Choose Meta Llama 3 8B Instruct for fits an 8 GB GPU and lighter on memory; choose GPT OSS 20B for higher llmrun Score and stronger at reasoning.
Which needs less VRAM, Meta Llama 3 8B Instruct or GPT OSS 20B?
At Q4_K_M, Meta Llama 3 8B Instruct requires 5.3 GB while GPT OSS 20B needs 12.9 GB, making Meta Llama 3 8B Instruct the easier model to run on your local setup.
Which is faster on RTX 4090?
Estimated decode speed on RTX 4090 is about 124 tok/s for Meta Llama 3 8B Instruct (Q4_K_M) and 190 tok/s for GPT OSS 20B (Q4_K_M), so GPT OSS 20B is faster. Estimates, ±20%.
Can I run Meta Llama 3 8B Instruct or GPT OSS 20B on a 16 GB GPU?
Meta Llama 3 8B Instruct fits at Q4_K_M; GPT OSS 20B fits at Q4_K_M.