Model comparison
GPT-4o vs Llama 3-8B
GPT-4o is the stronger model overall, scoring 28.5 to 25.3 on the Noometry Index. Llama 3-8B costs 31× less per token, which makes it the better buy when GPT-4o's lead doesn't matter for your workload.
Last verified . 29 shared benchmarks.
MetaSummary
- They share 29 benchmarks with published results for both. GPT-4o scores higher in 6 categories and Llama 3-8B in 2 categories; 8 gaps are clear of the uncertainty.
- The widest gap is in knowledge, where GPT-4o leads 28.8 to 7.8.
- The biggest single-benchmark swing is MATH Level 5: 53.3% for GPT-4o and 6.1% for Llama 3-8B.
- Llama 3-8B is cheaper at $0.14 / $0.14 per million input/output tokens, against $2.50 / $10 for GPT-4o.
- GPT-4o accepts more context: 128K tokens versus 8K.
- Llama 3-8B has downloadable open weights; the other is API-only.
Side by side
| GPT-4o | Llama 3-8B | |
|---|---|---|
| Provider | OpenAI | Meta |
| Noometry Index | 28.5 | 25.3 |
| Released | 2024-05-13 | 2024-04-18 |
| Weights | Proprietary | Open |
| Context window | 128K | 8K |
| Max output | 16K | 8K |
| Input $ / M tokens | $2.50 | $0.14 |
| Output $ / M tokens | $10 | $0.14 |
| Results tracked | 72 | 34 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Llama 3-8B leads
GPT-4o: 25.1 (#323), Llama 3-8B: 30.5 (#285)
| Benchmark | GPT-4o | Llama 3-8B |
|---|---|---|
| BigCodeBench Instruct | 51.1% | 31.9% |
| LMArena Coding | 1297 | 1152 |
| BigCodeBench Complete | 61.1% | 36.9% |
| HumanEval+ | 87.2% | 56.7% |
| MBPP+ | 72.2% | 54.8% |
| SWE-bench Verified | 31% | — |
| SWE-bench Verified (bash only) | 21.6% | — |
| Aider Polyglot | 45.3% | — |
| GSO | 0% | — |
| WeirdML | 25.1% | — |
| LiveBench Coding | 51.4% | — |
| CadEval | 26% | — |
Agentic & Tool Use Not comparable
GPT-4o: 20.8 (#133), Llama 3-8B: —
| Benchmark | GPT-4o | Llama 3-8B |
|---|---|---|
| GDPval | 9.9% | — |
| TheAgentCompany | 8.6% | — |
| Cybench | 12.5% | — |
| BALROG | 32.3% | — |
| LMArena Search | 1006 | — |
| METR Time Horizons | 40.8% | — |
Reasoning Llama 3-8B leads
GPT-4o: 9.4 (#338), Llama 3-8B: 13.8 (#321)
| Benchmark | GPT-4o | Llama 3-8B |
|---|---|---|
| Chess Puzzles | 13% | 0% |
| LMArena Hard Prompts | 1281 | 1133 |
| DTBench | 64.5% | 43.9% |
| Epoch Capabilities Index | 128.97 | 116.33 |
| ForecastBench | 57.7 | 58.6 |
| ARC-AGI-2 | 0% | — |
| SimpleBench | 17.8% | — |
| ARC-AGI-1 | 4.5% | — |
| CritPt | 0% | — |
| EnigmaEval | 0.8% | — |
| LiveBench Reasoning | 55.8% | — |
| LiveBench Data Analysis | 60.9% | — |
| LMCA | 16.6% | — |
| Adversarial NLI | — | 57.3% |
| LiveBench | 55.3% | — |
| WinoGrande | — | 75.7% |
Math GPT-4o leads
GPT-4o: 10.2 (#303), Llama 3-8B: 8.3 (#317)
| Benchmark | GPT-4o | Llama 3-8B |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 6.4% | 1.9% |
| LMArena Math | 1285 | 1151 |
| MATH Level 5 | 53.3% | 6.1% |
| FrontierMath (Tiers 1-3) | 0.4% | — |
| Omni-MATH | 29.3% | — |
| LiveBench Math | 49.5% | — |
| FrontierMath (Feb 2025 set) | 0.3% | — |
Knowledge GPT-4o leads
GPT-4o: 28.8 (#235), Llama 3-8B: 7.8 (#303)
| Benchmark | GPT-4o | Llama 3-8B |
|---|---|---|
| GPQA Diamond | 49.2% | 26.1% |
| LMArena Expert | 1250 | 1113 |
| MMLU | 88.1% | 68.8% |
| Humanity's Last Exam | 2.7% | — |
| SimpleQA Verified | 26% | — |
| MMLU-Pro | 71.3% | — |
| Confabulations | 15.3% | — |
| Vectara Hallucination Rate | 9.6% | — |
| GPQA (HELM) | 52% | — |
| ARC (AI2) Challenge | — | 82.8% |
| OpenBookQA | — | 82.6% |
| TriviaQA | — | 67.7% |
Multimodal Not comparable
GPT-4o: 33.9 (#85), Llama 3-8B: —
| Benchmark | GPT-4o | Llama 3-8B |
|---|---|---|
| LMArena Vision | 1137 | — |
| Video-MME | 71.9% | — |
| GeoBench | 71% | — |
| VPCT | 40% | — |
| ScienceQA | 88.5% | — |
Multilingual GPT-4o leads
GPT-4o: 43.4 (#181), Llama 3-8B: 31.0 (#256)
| Benchmark | GPT-4o | Llama 3-8B |
|---|---|---|
| LMArena Non-English | 1283 | 1098 |
| LMArena Chinese | 1277 | 1076 |
| LMArena French | 1303 | 1158 |
| LMArena German | 1282 | 1104 |
| LMArena Japanese | 1257 | 967 |
| LMArena Korean | 1234 | 1003 |
| LMArena Russian | 1286 | 1109 |
| LMArena Spanish | 1292 | 1172 |
Instruction Following GPT-4o leads
GPT-4o: 66.8 (#202), Llama 3-8B: 58.7 (#255)
| Benchmark | GPT-4o | Llama 3-8B |
|---|---|---|
| LMArena Instruction Following | 1278 | 1127 |
| LiveBench Instruction Following | 68.6% | — |
| IFEval | 81.7% | — |
Long Context GPT-4o leads
GPT-4o: 39.4 (#174), Llama 3-8B: 34.2 (#246)
| Benchmark | GPT-4o | Llama 3-8B |
|---|---|---|
| LMArena Longer Query | 1289 | 1128 |
| Fiction.LiveBench | 66.7% | — |
Writing & Preference GPT-4o leads
GPT-4o: 52.8 (#160), Llama 3-8B: 37.7 (#251)
| Benchmark | GPT-4o | Llama 3-8B |
|---|---|---|
| LMArena Text | 1300 | 1166 |
| LMArena Creative Writing | 1291 | 1150 |
| LMArena Multi-Turn | 1302 | 1152 |
| Short-Story Creative Writing | 81.8% | — |
| WildBench | 82.8% | — |
| LiveBench Language | 47.6% | — |
Frequently asked questions
Is GPT-4o better than Llama 3-8B?
GPT-4o is the stronger model overall, scoring 28.5 to 25.3 on the Noometry Index. Llama 3-8B costs 31× less per token, which makes it the better buy when GPT-4o's lead doesn't matter for your workload.
Which is cheaper, GPT-4o or Llama 3-8B?
Llama 3-8B is cheaper. It lists at $0.14 per million input tokens and $0.14 per million output tokens; GPT-4o lists at $2.50 and $10.
Is GPT-4o or Llama 3-8B better for coding?
Llama 3-8B scores higher on coding benchmarks: 30.5 versus 25.1 in the Noometry coding category.
Which has the bigger context window?
GPT-4o does, with 128K tokens against 8K.
How many benchmarks do GPT-4o and Llama 3-8B share?
29 benchmarks have published results for both models. GPT-4o has 72 scored results on Noometry and Llama 3-8B has 34.