Affärslivet

AI Model Profile · Meta AI

Llama 3.1-405B

Llama 3.1-405B is an AI model developed by Meta AI (United States), released July 23, 2024. It has approximately 405B parameters. It was trained on about 3.8×1025 FLOP of compute. Its weights are openly released.

Frontier model · Open weights · Epoch confidence: Confident · Source: Epoch AI · Updated 2026-08-17


Specifications

DeveloperMeta AI
CountryUnited States
ReleasedJuly 23, 2024
DomainLanguage
Parameters405B
Training compute3.8×1025 FLOP
Training cost (2023 USD)$53m
Training hardware16,384× NVIDIA H100 SXM5 80GB
Training power draw22.6 MW
WeightsOpen
AccessOpen weights (restricted use)
Frontier modelYes

All figures from Epoch AI's Notable AI Models database (CC BY 4.0). Blank fields are undisclosed by the developer — never estimated by Affärslivet.

Benchmarks

Performance on Epoch AI's reproducible evaluations (higher is better; scores are the best across scorers). These are independent runs, not developer-reported numbers.

AIME (OTIS Mock)9.7%
GPQA Diamond50.9%
MATH Level 549.8%

Source: Epoch AI Benchmarking Hub. GPQA Diamond = graduate-level science Q&A; SWE-Bench Verified = real software-engineering tasks; FrontierMath = research-level maths; MATH Level 5 = hardest competition maths; AIME = olympiad maths; SimpleQA = factual accuracy.

Frequently asked

Who developed Llama 3.1-405B?
Llama 3.1-405B was developed by Meta AI (United States) and released on July 23, 2024.
How big is Llama 3.1-405B?
Llama 3.1-405B has approximately 405B parameters and was trained on about 3.8×10^25 FLOP of compute, at an estimated training-compute cost of $53m. Source: Epoch AI.
Is Llama 3.1-405B open source?
Llama 3.1-405B is released with open weights (Yes), meaning its trained parameters are publicly available.
How does Llama 3.1-405B perform on benchmarks?
On Epoch AI's evaluations, Llama 3.1-405B scores 9.7% on AIME (OTIS Mock), 50.9% on GPQA Diamond, 49.8% on MATH Level 5. Source: Epoch AI Benchmarking Hub.

Sources

Cite this model profile

Free to cite, quote and reference under CC BY 4.0 — with attribution to Affärslivet. Writing an article or answer? Reference this model profile as:

APA

Affärslivet Research. (2026). Llama 3.1-405B — specifications & benchmarks. Affärslivet. https://xn--affrslivet-s5a.com/en/ai/models/llama-3-1-405b

MLA

"Llama 3.1-405B — specifications & benchmarks." Affärslivet, 2026-08-17, xn--affrslivet-s5a.com/en/ai/models/llama-3-1-405b.

Source: Affärslivetxn--affrslivet-s5a.com/en/ai/models/llama-3-1-405b. Attribute to Affärslivet when citing or linking.

Other AI models

← All AI models