Model comparison
Llama 3.1-8B vs Llama 3.2 3B
Llama 3.2 3B is the stronger model overall, scoring 28.4 to 22.8 on the Noometry Index. Llama 3.1-8B costs 2.1× less per token, which makes it the better buy when Llama 3.2 3B's lead doesn't matter for your workload.
Last verified . 18 shared benchmarks.
MetaSummary
- They share 18 benchmarks with published results for both. Llama 3.1-8B scores higher in 5 categories and Llama 3.2 3B in 4 categories; 9 gaps are clear of the uncertainty.
- The widest gap is in knowledge, where Llama 3.2 3B leads 29.1 to 8.0.
- The biggest single-benchmark swing is BigCodeBench Complete: 40.5% for Llama 3.1-8B and 28.3% for Llama 3.2 3B.
- Llama 3.1-8B is cheaper at $0.05 / $0.08 per million input/output tokens, against $0.05 / $0.33 for Llama 3.2 3B.
- Llama 3.2 3B accepts more context: 131K tokens versus 128K.
Side by side
| Llama 3.1-8B | Llama 3.2 3B | |
|---|---|---|
| Provider | Meta | Meta |
| Noometry Index | 22.8 | 28.4 |
| Released | 2024-07-23 | 2024-09-24 |
| Weights | Open | Open |
| Context window | 128K | 131K |
| Max output | 4K | 118K |
| Input $ / M tokens | $0.05 | $0.05 |
| Output $ / M tokens | $0.08 | $0.33 |
| Results tracked | 43 | 18 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Llama 3.2 3B leads
Llama 3.1-8B: 19.9 (#335), Llama 3.2 3B: 27.1 (#314)
| Benchmark | Llama 3.1-8B | Llama 3.2 3B |
|---|---|---|
| BigCodeBench Instruct | 32.8% | 23.4% |
| LMArena Coding | 1195 | 1098 |
| BigCodeBench Complete | 40.5% | 28.3% |
| SciCode | 13.2% | — |
| WeirdML | 1.7% | — |
| HumanEval+ | 62.8% | — |
| MBPP+ | 55.6% | — |
Agentic & Tool Use Llama 3.1-8B leads
Llama 3.1-8B: 22.3 (#123), Llama 3.2 3B: 19.9 (#135)
| Benchmark | Llama 3.1-8B | Llama 3.2 3B |
|---|---|---|
| Berkeley Function Calling Leaderboard | 25.8% | 21.9% |
| BALROG | 15.1% | 10.1% |
Reasoning Llama 3.2 3B leads
Llama 3.1-8B: 14.5 (#316), Llama 3.2 3B: 20.1 (#225)
| Benchmark | Llama 3.1-8B | Llama 3.2 3B |
|---|---|---|
| LMArena Hard Prompts | 1175 | 1095 |
| CritPt | 0% | — |
| Chess Puzzles | 0% | — |
| DTBench | 50.9% | — |
| LMCA | 5.4% | — |
| Epoch Capabilities Index | 116.48 | — |
| PIQA | 81.2% | — |
Math Llama 3.2 3B leads
Llama 3.1-8B: 9.5 (#311), Llama 3.2 3B: 29.9 (#213)
| Benchmark | Llama 3.1-8B | Llama 3.2 3B |
|---|---|---|
| LMArena Math | 1179 | 1126 |
| OTIS Mock AIME 2024-2025 | 1.7% | — |
| Omni-MATH | 13.7% | — |
| MATH Level 5 | 22.9% | — |
| GSM8K | 82.4% | — |
Knowledge Llama 3.2 3B leads
Llama 3.1-8B: 8.0 (#302), Llama 3.2 3B: 29.1 (#232)
| Benchmark | Llama 3.1-8B | Llama 3.2 3B |
|---|---|---|
| LMArena Expert | 1144 | 1090 |
| GPQA Diamond | 27% | — |
| MMLU-Pro | 40.6% | — |
| GPQA (HELM) | 24.7% | — |
| BoolQ | 82.8% | — |
| MMLU | 56.1% | — |
Multilingual Llama 3.1-8B leads
Llama 3.1-8B: 34.2 (#244), Llama 3.2 3B: 26.4 (#276)
| Benchmark | Llama 3.1-8B | Llama 3.2 3B |
|---|---|---|
| LMArena Non-English | 1148 | 1019 |
| LMArena Chinese | 1151 | 1017 |
| LMArena German | 1144 | 1056 |
| LMArena Russian | 1158 | 949 |
| LMArena French | 1176 | — |
| LMArena Japanese | 1061 | — |
| LMArena Korean | 1053 | — |
| LMArena Spanish | 1169 | — |
Instruction Following Llama 3.1-8B leads
Llama 3.1-8B: 59.1 (#253), Llama 3.2 3B: 56.3 (#270)
| Benchmark | Llama 3.1-8B | Llama 3.2 3B |
|---|---|---|
| LMArena Instruction Following | 1159 | 1089 |
| IFEval | 74.3% | — |
Long Context Llama 3.1-8B leads
Llama 3.1-8B: 35.8 (#233), Llama 3.2 3B: 33.4 (#256)
| Benchmark | Llama 3.1-8B | Llama 3.2 3B |
|---|---|---|
| LMArena Longer Query | 1182 | 1101 |
Writing & Preference Llama 3.1-8B leads
Llama 3.1-8B: 29.9 (#285), Llama 3.2 3B: 24.9 (#302)
| Benchmark | Llama 3.1-8B | Llama 3.2 3B |
|---|---|---|
| LMArena Text | 1187 | 1110 |
| LMArena Creative Writing | 1154 | 1094 |
| EQ-Bench Creative Writing | 713 | 595 |
| LMArena Multi-Turn | 1172 | 1105 |
| WildBench | 68.7% | — |
Frequently asked questions
Is Llama 3.1-8B better than Llama 3.2 3B?
Llama 3.2 3B is the stronger model overall, scoring 28.4 to 22.8 on the Noometry Index. Llama 3.1-8B costs 2.1× less per token, which makes it the better buy when Llama 3.2 3B's lead doesn't matter for your workload.
Which is cheaper, Llama 3.1-8B or Llama 3.2 3B?
Llama 3.1-8B is cheaper. It lists at $0.05 per million input tokens and $0.08 per million output tokens; Llama 3.2 3B lists at $0.05 and $0.33.
Is Llama 3.1-8B or Llama 3.2 3B better for coding?
Llama 3.2 3B scores higher on coding benchmarks: 27.1 versus 19.9 in the Noometry coding category.
Which has the bigger context window?
Llama 3.2 3B does, with 131K tokens against 128K.
How many benchmarks do Llama 3.1-8B and Llama 3.2 3B share?
18 benchmarks have published results for both models. Llama 3.1-8B has 43 scored results on Noometry and Llama 3.2 3B has 18.