It gave a broad group of developers access to capable models with commercial-use permissions subject to license conditions. Its 70B model competed with ChatGPT on many tasks, while coding remained a weakness for the general model.
Model size and corpus size are specifications, not quality scores. Comparisons in the report used different tasks and evaluators. Llama’s availability was subject to its license; small-model findings were limited to the evaluated settings.
Evidence you can use
Historical snapshot: October 2023. Dates and populations are specified per row.
| Measure | Reported value | Definition and source |
|---|---|---|
| Llama 2 pretraining corpus | 2T tokens | Corpus size reported for Llama 2, a 40% increase over its predecessor.2023 report, slide 19 (PDF page 19) |
| Llama 2 largest model discussed | 70B parameters | Model compared with ChatGPT across tasks in the report.2023 report, slide 19 (PDF page 19) |
| phi-1 model size | 1.3B parameters | Small code model trained on curated code and synthetic educational material.2023 report, slide 26 (PDF page 26) |
Model size and corpus size are specifications, not quality scores. Comparisons in the report used different tasks and evaluators. Llama’s availability was subject to its license; small-model findings were limited to the evaluated settings.
Historical snapshot published October 12, 2023. This web edition was prepared on October 10, 2026 from the online deck and original launch posts. Findings and forecasts retain their original time frame.
Benaich, Nathan. “What did Llama 2 change in 2023?.” State of AI Report 2023. Historical report snapshot; web edition prepared 2026-10-10.