Meta Expands Open-Source Push with Llama 3.1 405B and Sub-Cent 8B API Pricing

Meta introduced Llama 3.1 405B, which expands context length to 128K and adds support across eight languages. [1] The ecosystem includes over 25 partners, including AWS, NVIDIA, Databricks, and Google Cloud, offering services on day one. [1]

Meta also announced new security and safety tools, including Llama Guard 3 and Prompt Guard, and released a request for comment on the Llama Stack API. [1]

For API access, the text-only Llama 3 8B model features a standard context window of 8,192 tokens. [2] The Llama 3 8B model is priced competitively for developers at $0.050 per million input tokens, while generation costs are set at $0.080 per million output tokens. [2]

Meta is aggressively targeting both ends of the AI ecosystem by releasing the 405B model while maintaining low operational costs on its smaller architectures. The 405B model introduces a massive 128K context window that positions Meta's open-source foundation model directly against leading closed alternatives. At the same time, the Llama 3 8B model competes heavily on price at $0.050 per million input tokens, positioning it tightly against other small models like Google's Gemma 3 4B, which prices input tokens at $0.040 per million. What the sources don't address: whether Meta plans to introduce multi-modal capabilities to the highly affordable 8B model.