Llama 3.1 70B
General language model Generally available Open WeightsMeta AI open-weights 70-billion parameter model balancing high-tier language intelligence with local GPU server deployment feasibility.
Overview
Llama 3.1 70B is Meta AI's mid-sized open-weights language model released in July 2024. Designed as the optimal balance between high intelligence and efficient infrastructure deployment, 70B provides enterprise-grade performance for coding, complex reasoning, RAG pipelines, and conversational agents.
The model features a 131,072-token (128K) context window, expanding drastically over prior Llama 3 releases. Trained on 15+ trillion tokens with Grouped-Query Attention, it delivers state-of-the-art open-weights performance. It supports tool calling, JSON outputs, and multilingual processing across 8 officially supported languages.
Released under the Llama 3.1 Community License, weights are available on Hugging Face and cloud platforms. Local deployment is widely supported across Ollama, vLLM, LM Studio, and local workstation hardware.
Technical Specifications
Context Window
131,072 tokens
Max Input Tokens
128,000
Parameters
70.0B
Architecture
Dense Autoregressive Transformer
Released
2024-07-23
Modalities
Text → Text (Both)
Capabilities
Text Generation
Code Generation
Reasoning
Function Calling
Long Context
Use Cases
Coding Assistance
Research
Languages
English (en)
Spanish (es)
French (fr)
German (de)
Access Routes
| Route | Type | Protocol | Region / Scope | Provider | Status |
|---|---|---|---|---|---|
| Meta AI | Hosted API (Direct provider) | OpenAI-compatible REST API | Global | Meta AI | Available |
Identifiers
- meta-llama/Meta-Llama-3.1-70B-Instruct Canonical identifier Default
- meta-llama/Llama-3.1-70B-Instruct Canonical identifier Default
- meta-llama/Llama-3.1-70B Alias