Model comparison
GPT-4 vs Llama-3.3-70B-Instruct
Llama-3.3-70B-Instruct is the stronger model overall, scoring 30.6 to 29.1 on the Noometry Index.
Last verified . 28 shared benchmarks.
MetaSummary
- They share 28 benchmarks with published results for both. GPT-4 scores higher in 4 categories and Llama-3.3-70B-Instruct in 4 categories; 6 gaps are clear of the uncertainty.
- The widest gap is in writing & preference, where Llama-3.3-70B-Instruct leads 47.6 to 34.9.
- The biggest single-benchmark swing is MATH Level 5: 23% for GPT-4 and 41.6% for Llama-3.3-70B-Instruct.
- Llama-3.3-70B-Instruct is cheaper at $0.10 / $0.32 per million input/output tokens, against $30 / $60 for GPT-4.
- Llama-3.3-70B-Instruct accepts more context: 128K tokens versus 8K.
- Llama-3.3-70B-Instruct has downloadable open weights; the other is API-only.
Side by side
| GPT-4 | Llama-3.3-70B-Instruct | |
|---|---|---|
| Provider | OpenAI | Meta |
| Noometry Index | 29.1 | 30.6 |
| Released | 2023-03-14 | 2024-12-06 |
| Weights | Proprietary | Open |
| Context window | 8K | 128K |
| Max output | 8K | 4K |
| Input $ / M tokens | $30 | $0.10 |
| Output $ / M tokens | $60 | $0.32 |
| Results tracked | 38 | 43 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Too close to call
GPT-4: 31.6 (#283), Llama-3.3-70B-Instruct: 31.0 (#290)
| Benchmark | GPT-4 | Llama-3.3-70B-Instruct |
|---|---|---|
| WeirdML | 12.4% | 14.4% |
| BigCodeBench Instruct | 46% | 46.9% |
| LMArena Coding | 1254 | 1268 |
| BigCodeBench Complete | 57.2% | 57.5% |
| SciCode | — | 26% |
| LiveBench Coding | — | 36.6% |
| HumanEval+ | 79.3% | — |
Agentic & Tool Use Not comparable
GPT-4: —, Llama-3.3-70B-Instruct: 25.8 (#105)
| Benchmark | GPT-4 | Llama-3.3-70B-Instruct |
|---|---|---|
| Berkeley Function Calling Leaderboard | — | 31.9% |
| BALROG | — | 23% |
| METR Time Horizons | 36.1% | — |
Reasoning GPT-4 leads
GPT-4: 17.8 (#289), Llama-3.3-70B-Instruct: 14.1 (#327)
| Benchmark | GPT-4 | Llama-3.3-70B-Instruct |
|---|---|---|
| LMArena Hard Prompts | 1241 | 1257 |
| DTBench | 62.7% | 59.5% |
| LMCA | 17.1% | 17.5% |
| Epoch Capabilities Index | 125.89 | 127.33 |
| ForecastBench | 57.8 | 58.6 |
| SimpleBench | — | 19.9% |
| CritPt | — | 0% |
| Chess Puzzles | 4% | — |
| LiveBench Reasoning | — | 50.8% |
| Mystery Game Puzzles | 12% | — |
| LiveBench Data Analysis | — | 49.5% |
| BIG-Bench Hard | 75.1% | — |
| HellaSwag | 95.3% | — |
| LiveBench | — | 50.2% |
| WinoGrande | 87.5% | — |
Math Llama-3.3-70B-Instruct leads
GPT-4: 10.8 (#309), Llama-3.3-70B-Instruct: 15.3 (#298)
| Benchmark | GPT-4 | Llama-3.3-70B-Instruct |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 1.1% | 5.1% |
| LMArena Math | 1269 | 1267 |
| MATH Level 5 | 23% | 41.6% |
| LiveBench Math | — | 42.2% |
| GSM8K | 92% | — |
Knowledge Llama-3.3-70B-Instruct leads
GPT-4: 18.4 (#282), Llama-3.3-70B-Instruct: 30.6 (#226)
| Benchmark | GPT-4 | Llama-3.3-70B-Instruct |
|---|---|---|
| GPQA Diamond | 35.7% | 47.4% |
| LMArena Expert | 1211 | 1225 |
| MMLU | 86.4% | 86.3% |
| Confabulations | — | 22.8% |
| Vectara Hallucination Rate | — | 4.1% |
| TriviaQA | 84.8% | — |
Multilingual Too close to call
GPT-4: 40.6 (#215), Llama-3.3-70B-Instruct: 39.9 (#220)
| Benchmark | GPT-4 | Llama-3.3-70B-Instruct |
|---|---|---|
| LMArena Non-English | 1246 | 1236 |
| LMArena Chinese | 1242 | 1217 |
| LMArena French | 1283 | 1281 |
| LMArena German | 1251 | 1251 |
| LMArena Japanese | 1209 | 1150 |
| LMArena Korean | 1184 | 1143 |
| LMArena Russian | 1251 | 1252 |
| LMArena Spanish | 1261 | 1270 |
Instruction Following Llama-3.3-70B-Instruct leads
GPT-4: 65.3 (#222), Llama-3.3-70B-Instruct: 71.1 (#157)
| Benchmark | GPT-4 | Llama-3.3-70B-Instruct |
|---|---|---|
| LMArena Instruction Following | 1241 | 1242 |
| LiveBench Instruction Following | — | 82.7% |
Long Context GPT-4 leads
GPT-4: 37.7 (#212), Llama-3.3-70B-Instruct: 26.4 (#295)
| Benchmark | GPT-4 | Llama-3.3-70B-Instruct |
|---|---|---|
| LMArena Longer Query | 1244 | 1256 |
| Fiction.LiveBench | — | 33.3% |
Writing & Preference Llama-3.3-70B-Instruct leads
GPT-4: 34.9 (#268), Llama-3.3-70B-Instruct: 47.6 (#207)
| Benchmark | GPT-4 | Llama-3.3-70B-Instruct |
|---|---|---|
| LMArena Text | 1263 | 1274 |
| LMArena Creative Writing | 1244 | 1250 |
| LMArena Multi-Turn | 1257 | 1280 |
| EQ-Bench Creative Writing | 752 | — |
| LiveBench Language | — | 39.2% |
Frequently asked questions
Is GPT-4 better than Llama-3.3-70B-Instruct?
Llama-3.3-70B-Instruct is the stronger model overall, scoring 30.6 to 29.1 on the Noometry Index.
Which is cheaper, GPT-4 or Llama-3.3-70B-Instruct?
Llama-3.3-70B-Instruct is cheaper. It lists at $0.10 per million input tokens and $0.32 per million output tokens; GPT-4 lists at $30 and $60.
Is GPT-4 or Llama-3.3-70B-Instruct better for coding?
They score almost the same on coding (31.6 vs 31.0); test both on your own repository before choosing.
Which has the bigger context window?
Llama-3.3-70B-Instruct does, with 128K tokens against 8K.
How many benchmarks do GPT-4 and Llama-3.3-70B-Instruct share?
28 benchmarks have published results for both models. GPT-4 has 38 scored results on Noometry and Llama-3.3-70B-Instruct has 43.