Meta AI

Llama 3.1 405B Instruct

Type
language model
Context
128K tokens
Max output
128K
Released
23 July 2024
API string
llama-3.1-405b-instruct
Weights
open

Prices by provider

Provider Input, $ per 1M Output, $ per 1M In our data since
Hyperbolic $0.12 $0.3 1 Aug 2026
Nebius Token Factory $1 $3 1 Aug 2026
Google Vertex AI $5 $16 31 Jul 2026
SambaNova $5 $10 1 Aug 2026
Databricks $5 $15 1 Aug 2026
Amazon Bedrock $5.32 $16 2 Aug 2026
Microsoft Foundry $5.33 $16 1 Aug 2026
OCI Generative AI $10.68 $10.68 31 Jul 2026

Only the base tier and only per-token prices are shown. Batch, discounted and cached rates, as well as prices per image or per second of video, do not go into this table: they cannot stand in the same column as a price per million tokens.

The date in the last column is the day this price first entered our collection. The price may well be older: before that day we simply were not recording it. It has not changed since — otherwise a new row with a new date would stand in its place.

Measurement results

Task set Result Run conditions Measured by
Arena Score in English 1,308.06 1,303–1,313
Arena Score, multi-turn dialogue 1,287.34 1,280–1,295
Arena Score, programming 1,284 1,277–1,291
Arena Score, overall 1,281.97 1,278–1,286
Arena Score, mathematics 1,278.23 1,270–1,286
Arena Score in French 1,269.69 1,242–1,297
Arena Score, hard prompts 1,263.75 1,258–1,270
Arena Score, creative writing 1,260.14 1,253–1,268
Arena Score, long queries 1,259.79 1,252–1,267
Arena Score, instruction following 1,259.06 1,254–1,264
Arena Score in Russian 1,255.23 1,247–1,264
Arena Score in Spanish 1,252.29 1,227–1,277
Arena Score, expert questions 1,228.55 1,216–1,241
MATH, difficulty level five 49.77 % · with a tuned harness Epoch evaluations
GPQA Diamond — graduate-level questions 34.55 % · with a tuned harness Epoch evaluations
WeirdML — unusual machine learning tasks 21.38 % WeirdML Leaderboard
Mock AIME 2024–2025 — olympiad problems 9.63 % · with a tuned harness Epoch evaluations
SimpleBench — trick questions 7.6 % SimpleBench Leaderboard
Cybench — cybersecurity tasks 7.5 % · with a tuned harness Cybench leaderboard
The Agent Company — work tasks in an office environment 7.4 % TheAgentCompany experiment results github