Model comparison
GPT-5.4 mini vs Llama 3.2 3B
GPT-5.4 mini is the stronger model overall, scoring 45.0 to 28.9 on the Noometry Index. Llama 3.2 3B costs 14× less per token, which makes it the better buy when GPT-5.4 mini's lead doesn't matter for your workload.
Last verified . 14 shared benchmarks.
MetaSummary
- They share 14 benchmarks with published results for both. GPT-5.4 mini scores higher in 9 categories and Llama 3.2 3B in 0 categories; 9 gaps are clear of the uncertainty.
- The widest gap is in writing & preference, where GPT-5.4 mini leads 64.0 to 24.7.
- Llama 3.2 3B is cheaper at $0.05 / $0.33 per million input/output tokens, against $0.75 / $4.50 for GPT-5.4 mini.
- GPT-5.4 mini accepts more context: 400K tokens versus 131K.
- Llama 3.2 3B has downloadable open weights; the other is API-only.
Side by side
| GPT-5.4 mini | Llama 3.2 3B | |
|---|---|---|
| Provider | OpenAI | Meta |
| Noometry Index | 45.0 | 28.9 |
| Released | 2026-03-17 | 2024-09-24 |
| Weights | Proprietary | Open |
| Context window | 400K | 131K |
| Max output | 128K | 118K |
| Input $ / M tokens | $0.75 | $0.05 |
| Output $ / M tokens | $4.50 | $0.33 |
| Results tracked | 46 | 18 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding GPT-5.4 mini leads
GPT-5.4 mini: 45.2 (#72), Llama 3.2 3B: 27.6 (#319)
| Benchmark | GPT-5.4 mini | Llama 3.2 3B |
|---|---|---|
| LMArena Coding | 1438 | 1098 |
| FrontierCode | 27% | — |
| LMArena WebDev | 1397 | — |
| SciCode | 49.9% | — |
| WeirdML | 60.3% | — |
| BigCodeBench Instruct | — | 23.4% |
| BigCodeBench Complete | — | 28.3% |
| ALE-Bench | 1,189 | — |
Agentic & Tool Use GPT-5.4 mini leads
GPT-5.4 mini: 29.9 (#81), Llama 3.2 3B: 20.1 (#143)
| Benchmark | GPT-5.4 mini | Llama 3.2 3B |
|---|---|---|
| Berkeley Function Calling Leaderboard | — | 21.9% |
| DeepResearch Bench | 36.3% | — |
| BALROG | — | 10.1% |
Reasoning GPT-5.4 mini leads
GPT-5.4 mini: 30.4 (#85), Llama 3.2 3B: 21.0 (#228)
| Benchmark | GPT-5.4 mini | Llama 3.2 3B |
|---|---|---|
| LMArena Hard Prompts | 1424 | 1095 |
| ARC-AGI-2 | 18.9% | — |
| Kagi LLM Benchmark | 37.9% | — |
| NYT Connections (extended) | 61.8% | — |
| ARC-AGI-1 | 63.7% | — |
| CritPt | 10% | — |
| Chess Puzzles | 24% | — |
| Thematic Generalization | 61.7% | — |
| Mystery Game Puzzles | 11% | — |
| DTBench | 80% | — |
| LMCA | 40.8% | — |
| Epoch Capabilities Index | 148.84 | — |
| ForecastBench | 57 | — |
Math GPT-5.4 mini leads
GPT-5.4 mini: 45.5 (#75), Llama 3.2 3B: 32.4 (#214)
| Benchmark | GPT-5.4 mini | Llama 3.2 3B |
|---|---|---|
| LMArena Math | 1419 | 1126 |
| FrontierMath (Tiers 1-3) | 51.2% | — |
| FrontierMath Tier 4 | 9.8% | — |
| OTIS Mock AIME 2024-2025 | 88.9% | — |
| ProofBench | 21% | — |
| FrontierMath (Feb 2025 set) | 28.3% | — |
| FrontierMath Tier 4 (v1) | 2.1% | — |
Knowledge GPT-5.4 mini leads
GPT-5.4 mini: 51.5 (#67), Llama 3.2 3B: 29.7 (#235)
| Benchmark | GPT-5.4 mini | Llama 3.2 3B |
|---|---|---|
| LMArena Expert | 1435 | 1090 |
| GPQA Diamond | 86.9% | — |
| SimpleQA Verified | 29.4% | — |
| Vectara Hallucination Rate | 5.5% | — |
Multimodal Not comparable
GPT-5.4 mini: 39.7 (#56), Llama 3.2 3B: —
| Benchmark | GPT-5.4 mini | Llama 3.2 3B |
|---|---|---|
| LMArena Vision | 1245 | — |
Multilingual GPT-5.4 mini leads
GPT-5.4 mini: 51.9 (#96), Llama 3.2 3B: 26.2 (#281)
| Benchmark | GPT-5.4 mini | Llama 3.2 3B |
|---|---|---|
| LMArena Non-English | 1405 | 1019 |
| LMArena Chinese | 1446 | 1017 |
| LMArena German | 1409 | 1056 |
| LMArena Russian | 1417 | 949 |
| LMArena French | 1440 | — |
| LMArena Japanese | 1374 | — |
| LMArena Korean | 1368 | — |
| LMArena Spanish | 1405 | — |
Instruction Following GPT-5.4 mini leads
GPT-5.4 mini: 74.1 (#102), Llama 3.2 3B: 56.0 (#275)
| Benchmark | GPT-5.4 mini | Llama 3.2 3B |
|---|---|---|
| LMArena Instruction Following | 1405 | 1089 |
Long Context GPT-5.4 mini leads
GPT-5.4 mini: 43.0 (#112), Llama 3.2 3B: 33.4 (#261)
| Benchmark | GPT-5.4 mini | Llama 3.2 3B |
|---|---|---|
| LMArena Longer Query | 1407 | 1100 |
Writing & Preference GPT-5.4 mini leads
GPT-5.4 mini: 64.0 (#58), Llama 3.2 3B: 24.7 (#307)
| Benchmark | GPT-5.4 mini | Llama 3.2 3B |
|---|---|---|
| LMArena Text | 1412 | 1110 |
| LMArena Creative Writing | 1370 | 1094 |
| EQ-Bench Creative Writing | 1665 | 595 |
| LMArena Multi-Turn | 1429 | 1105 |
Frequently asked questions
Is GPT-5.4 mini better than Llama 3.2 3B?
GPT-5.4 mini is the stronger model overall, scoring 45.0 to 28.9 on the Noometry Index. Llama 3.2 3B costs 14× less per token, which makes it the better buy when GPT-5.4 mini's lead doesn't matter for your workload.
Which is cheaper, GPT-5.4 mini or Llama 3.2 3B?
Llama 3.2 3B is cheaper. It lists at $0.05 per million input tokens and $0.33 per million output tokens; GPT-5.4 mini lists at $0.75 and $4.50.
Is GPT-5.4 mini or Llama 3.2 3B better for coding?
GPT-5.4 mini scores higher on coding benchmarks: 45.2 versus 27.6 in the Noometry coding category.
Which has the bigger context window?
GPT-5.4 mini does, with 400K tokens against 131K.
How many benchmarks do GPT-5.4 mini and Llama 3.2 3B share?
14 benchmarks have published results for both models. GPT-5.4 mini has 46 scored results on Noometry and Llama 3.2 3B has 18.