Inference
EN
Meta

Llama 3.1 405B

Llama 3.1 405B is a standard model from Meta. 9 benchmarks count toward its index, in 5 categories.

  • Open weights
  • Standard
  • Input: text
  • Released 2024-07
RankedEnough results for an index.
IndexAll categories, out of 100.27.0±8.9Median 51.0
CoverageIndex weight that has results.65%
SpeedOutput tokens per second.29/sMedian 67/s
First tokenSeconds to the first output token.2.2 sMedian 1.3 s
Input / 1MUS dollars per 1M input tokens.—Median $0.60
Output / 1MUS dollars per 1M output tokens.—Median $2.55
ContextMaximum tokens in one request.128KMedian 262K
EloLMArena rating and rank.—Median 1422

Run locally

405.85B parameters
16-bit
853.3 GB
8-bit
453.8 GB
4-bit
259.3 GB
Weights
Hugging Face (opens in a new tab) ↗

Memory need with a short context. The 4-bit copy needs more than 128 GB. Compare local models

Capabilities

Index per category, out of 100.
Out of 100
AgenticMulti-step tasks with tools.1 counted15.4—
CodingCode writing and repair.None counted——
ReasoningLogic problems and puzzles.2 counted30.8207 / 227
MultimodalTasks with images and text.None counted——
KnowledgeFacts and expert knowledge.2 counted26.3253 / 268
MultilingualTasks in many languages.None counted——
InstructionTasks with strict prompt rules.2 counted33.550 / 63
MathMath problems.2 counted18.9160 / 168

Results

9 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
Instruction-Following EvalInstruction88.6%30 / 70
MATH level 5Math49.8% ±1.249 / 88
GPQA diamond · Epoch AIKnowledge50.9% ±2.6143 / 217
Critical Physics TasksReasoning0.0%150 / 214
Artificial Analysis IFBenchInstruction39.0%108 / 146
OTIS Mock AIME 2024-2025Math9.7% ±3.2155 / 200
Artificial Analysis GPQA DiamondKnowledge51.5%170 / 199
Artificial Analysis Long Context ReasoningReasoning25.3%189 / 217
τ²-Bench Tool-Agent-User EvaluationAgentic19.0%134 / 149
Artificial Analysis Humanity's Last ExamKnowledge4.0%201 / 222

Sources

BenchLMCC BY-NC 4.0Epoch AICC BY

History

5 releases
Index28.7 → 27.0
23 Oct 202612 Nov 2026

Questions

How good is Llama 3.1 405B?
Its index is 27.0 out of 100. That is place 263 of 283 ranked models.
What is Llama 3.1 405B best at?
Instruction: 33.5 out of 100, place 50 of 63.
How fast is Llama 3.1 405B?
29 output tokens per second, 2.2 s to the first token.
What is the context window of Llama 3.1 405B?
128K tokens.
Is Llama 3.1 405B open source?
Yes. Its weights are published.
When was Llama 3.1 405B released?
July 2024.

Badge

Llama 3.1 405B on Inference360: place 263 of 283, index 27.0
HTML
<a href="https://inference360.ai/models/llama-3-1-405b"><img src="https://inference360.ai/badges/llama-3-1-405b.svg" alt="Llama 3.1 405B on Inference360: place 263 of 283, index 27.0" height="20"></a>
Markdown
[![Llama 3.1 405B on Inference360: place 263 of 283, index 27.0](https://inference360.ai/badges/llama-3-1-405b.svg)](https://inference360.ai/models/llama-3-1-405b)

Compare with

Phi-4-reasoning27.1Gemini 2.5 Flash-Lite27.1Llama 4 Scout26.8GPT-4.1 nano26.3Llama 3.1 70B24.8

More from Meta

All models
Llama 3.1 70B24.8Llama 3.1 8B17.9Llama 3.1 70B InstructUnrankedLlama 3.1 8B InstructUnrankedLlama 3.2 1B InstructUnranked