Model comparison
gpt-oss-20b vs Llama-3.3-70B-Instruct
gpt-oss-20b is the stronger model overall, scoring 32.5 to 30.6 on the Noometry Index.
Last verified . 24 shared benchmarks.
MetaSummary
- They share 24 benchmarks with published results for both. gpt-oss-20b scores higher in 6 categories and Llama-3.3-70B-Instruct in 3 categories; 9 gaps are clear of the uncertainty.
- The widest gap is in math, where gpt-oss-20b leads 39.4 to 15.3.
- The biggest single-benchmark swing is OTIS Mock AIME 2024-2025: 65.3% for gpt-oss-20b and 5.1% for Llama-3.3-70B-Instruct.
- gpt-oss-20b is cheaper at $0.018 / $0.09 per million input/output tokens, against $0.10 / $0.32 for Llama-3.3-70B-Instruct.
- gpt-oss-20b accepts more context: 131K tokens versus 128K.
Side by side
| gpt-oss-20b | Llama-3.3-70B-Instruct | |
|---|---|---|
| Provider | OpenAI | Meta |
| Noometry Index | 32.5 | 30.6 |
| Released | 2025-08-05 | 2024-12-06 |
| Weights | Open | Open |
| Context window | 131K | 128K |
| Max output | 16K | 4K |
| Input $ / M tokens | $0.018 | $0.10 |
| Output $ / M tokens | $0.09 | $0.32 |
| Results tracked | 34 | 43 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding gpt-oss-20b leads
gpt-oss-20b: 37.6 (#192), Llama-3.3-70B-Instruct: 31.0 (#290)
| Benchmark | gpt-oss-20b | Llama-3.3-70B-Instruct |
|---|---|---|
| SciCode | 34.4% | 26% |
| WeirdML | 40.9% | 14.4% |
| LMArena Coding | 1306 | 1268 |
| BigCodeBench Instruct | — | 46.9% |
| LiveBench Coding | — | 36.6% |
| BigCodeBench Complete | — | 57.5% |
| ALE-Bench | 566.05 | — |
Agentic & Tool Use Llama-3.3-70B-Instruct leads
gpt-oss-20b: 9.3 (#154), Llama-3.3-70B-Instruct: 25.8 (#105)
| Benchmark | gpt-oss-20b | Llama-3.3-70B-Instruct |
|---|---|---|
| Terminal-Bench | 3.4% | — |
| Berkeley Function Calling Leaderboard | — | 31.9% |
| BALROG | — | 23% |
Reasoning gpt-oss-20b leads
gpt-oss-20b: 19.3 (#261), Llama-3.3-70B-Instruct: 14.1 (#327)
| Benchmark | gpt-oss-20b | Llama-3.3-70B-Instruct |
|---|---|---|
| CritPt | 1.4% | 0% |
| LMArena Hard Prompts | 1274 | 1257 |
| DTBench | 68% | 59.5% |
| LMCA | 14.5% | 17.5% |
| Epoch Capabilities Index | 137.82 | 127.33 |
| SimpleBench | — | 19.9% |
| Kagi LLM Benchmark | 53.2% | — |
| Chess Puzzles | 4% | — |
| LiveBench Reasoning | — | 50.8% |
| LiveBench Data Analysis | — | 49.5% |
| ForecastBench | — | 58.6 |
| LiveBench | — | 50.2% |
Math gpt-oss-20b leads
gpt-oss-20b: 39.4 (#103), Llama-3.3-70B-Instruct: 15.3 (#298)
| Benchmark | gpt-oss-20b | Llama-3.3-70B-Instruct |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 65.3% | 5.1% |
| LMArena Math | 1317 | 1267 |
| Omni-MATH | 56.5% | — |
| LiveBench Math | — | 42.2% |
| MATH Level 5 | — | 41.6% |
Knowledge gpt-oss-20b leads
gpt-oss-20b: 34.6 (#195), Llama-3.3-70B-Instruct: 30.6 (#226)
| Benchmark | gpt-oss-20b | Llama-3.3-70B-Instruct |
|---|---|---|
| GPQA Diamond | 60.8% | 47.4% |
| LMArena Expert | 1258 | 1225 |
| MMLU-Pro | 74% | — |
| Confabulations | — | 22.8% |
| Vectara Hallucination Rate | — | 4.1% |
| GPQA (HELM) | 59.4% | — |
| MMLU | — | 86.3% |
Multilingual gpt-oss-20b leads
gpt-oss-20b: 42.2 (#197), Llama-3.3-70B-Instruct: 39.9 (#220)
| Benchmark | gpt-oss-20b | Llama-3.3-70B-Instruct |
|---|---|---|
| LMArena Non-English | 1268 | 1236 |
| LMArena Chinese | 1314 | 1217 |
| LMArena German | 1255 | 1251 |
| LMArena Japanese | 1244 | 1150 |
| LMArena Korean | 1236 | 1143 |
| LMArena Russian | 1278 | 1252 |
| LMArena Spanish | 1267 | 1270 |
| LMArena French | — | 1281 |
Instruction Following Llama-3.3-70B-Instruct leads
gpt-oss-20b: 61.8 (#240), Llama-3.3-70B-Instruct: 71.1 (#157)
| Benchmark | gpt-oss-20b | Llama-3.3-70B-Instruct |
|---|---|---|
| LMArena Instruction Following | 1236 | 1242 |
| LiveBench Instruction Following | — | 82.7% |
| IFEval | 73.2% | — |
Long Context gpt-oss-20b leads
gpt-oss-20b: 37.9 (#209), Llama-3.3-70B-Instruct: 26.4 (#295)
| Benchmark | gpt-oss-20b | Llama-3.3-70B-Instruct |
|---|---|---|
| LMArena Longer Query | 1250 | 1256 |
| Fiction.LiveBench | — | 33.3% |
Writing & Preference Llama-3.3-70B-Instruct leads
gpt-oss-20b: 35.5 (#265), Llama-3.3-70B-Instruct: 47.6 (#207)
| Benchmark | gpt-oss-20b | Llama-3.3-70B-Instruct |
|---|---|---|
| LMArena Text | 1287 | 1274 |
| LMArena Creative Writing | 1201 | 1250 |
| LMArena Multi-Turn | 1268 | 1280 |
| EQ-Bench Creative Writing | 666 | — |
| WildBench | 73.7% | — |
| LiveBench Language | — | 39.2% |
Frequently asked questions
Is gpt-oss-20b better than Llama-3.3-70B-Instruct?
gpt-oss-20b is the stronger model overall, scoring 32.5 to 30.6 on the Noometry Index.
Which is cheaper, gpt-oss-20b or Llama-3.3-70B-Instruct?
gpt-oss-20b is cheaper. It lists at $0.018 per million input tokens and $0.09 per million output tokens; Llama-3.3-70B-Instruct lists at $0.10 and $0.32.
Is gpt-oss-20b or Llama-3.3-70B-Instruct better for coding?
gpt-oss-20b scores higher on coding benchmarks: 37.6 versus 31.0 in the Noometry coding category.
Which has the bigger context window?
gpt-oss-20b does, with 131K tokens against 128K.
How many benchmarks do gpt-oss-20b and Llama-3.3-70B-Instruct share?
24 benchmarks have published results for both models. gpt-oss-20b has 34 scored results on Noometry and Llama-3.3-70B-Instruct has 43.