Models · Compare
Llama 3.2 3B Instruct vs Claude 3.5 Haiku
Rows highlighted in warm gray are where the models differ. Numbers carry their as-of date and primary source.
Specs
| Field | A: Llama 3.2 3B Instruct | B: Claude 3.5 Haiku |
|---|---|---|
| Released | 2024-09-25 | 2024-11-04 |
| Developer | Meta | Anthropic |
| Openness | Source-available | Proprietary |
| License | Llama 3.2 Community License | Proprietary |
| OSI-approved | no | no |
| Data released | no | no |
| Training code | no | no |
| Architecture | dense | unknown |
| Total params | 3.2B | — |
| Active params | — | — |
| Experts | — | — |
| Context window | 131K | 200K |
| Attention | gqa | unknown |
| Position enc. | rope-llama3 | unknown |
| Pretraining tokens | — | — |
| Post-training | sft, dpo, rejection-sampling, rlhf | rlhf, constitutional |
| Training hardware | H100 | — |
| $/M input | — | $0.80 |
| $/M output | — | $4.00 |
| Output tok/sec | — | 0 |
Benchmarks
Missing scores render as not reported; never inferred. Bold highlights the leader per benchmark.
General reasoning
| MMLU-Pro | — | 63.4 2026-05-21 |
| GPQA-Diamond | — | 40.8 2026-05-21 |
Code
| SWE-Bench Verified | — | 40.6 2024-10-22 |
| LiveCodeBench | — | 31.4 2026-05-21 |
Math
| MATH | — | 72.1 2026-05-21 |
| AIME 2024 | — | 3.3 2026-05-21 |
Held-out / arena
Context · A
The larger of the two text-only Llama 3.2 lightweight checkpoints, also produced by pruning and distilling from Llama 3.1 8B. Meta positioned the 3B as the on-device tier for summarization, rewriting, and tool-use tasks.
Context · B
Announced October 22 2024 alongside the upgraded Claude 3.5 Sonnet and the computer-use beta; generally available in early November. Anthropic positioned it as matching Claude 3 Opus on many intelligence benchmarks at Haiku-tier speeds, with a SWE-Bench Verified score of 40.6% making it briefly the best small-tier coding model from a major lab.