Meta released Llama 3.1 on July 23, 2024, as a family of three text-in/text-out large language models: 8B, 70B and 405B. The models were made available for download under Meta’s Llama 3.1 Community License, with a stated 128K-token context length and support for eight languages. Meta framed the largest model as a competitor to leading systems such as GPT-4o, but that was Meta’s launch-era experimental evaluation—not evidence of a current or across-the-board win.
What is Llama 3.1?
Llama 3.1 is Meta’s July 2024 release of a family of large language models. Meta described the collection as text-in/text-out, so it should not be treated as a multimodal model release. The official model card covers pretrained and instruction-tuned versions: instruction-tuned models are intended for multilingual dialogue, while pretrained models can be adapted to other natural-language-generation tasks. Meta’s launch announcement and official model card provide the release details.
Model sizes and context
| Variant | Parameters | Stated context length | What the size tells you |
|---|---|---|---|
| Llama 3.1 8B | 8 billion | 128K tokens, as stated by Meta in 2024 | Smallest member of the family; parameter count alone does not establish the hardware needed for a particular deployment. |
| Llama 3.1 70B | 70 billion | 128K tokens, as stated by Meta in 2024 | Middle-sized member of the family; compare its intended capability with the compute and latency your application can support. |
| Llama 3.1 405B | 405 billion | 128K tokens, as stated by Meta in 2024 | Largest member; Meta reported that it was trained on more than 15 trillion tokens. |
Meta listed English, German, French, Italian, Portuguese, Hindi, Spanish and Thai as supported languages. The model card gives December 2023 as the pretraining-data cutoff. These are launch-era specifications, not a statement about later updates or current service availability.
How did Meta say Llama 3.1 compared with OpenAI models?
Meta said it tested the release on more than 150 benchmark datasets spanning languages and also conducted human evaluations. Its July 23, 2024 announcement described the 405B model as competitive with GPT-4, GPT-4o and Claude 3.5 Sonnet across a range of tasks, and said the smaller variants were competitive with similarly sized models. Meta called this an “experimental evaluation”; it is the company’s reported launch-era result, not an independent ranking or a current comparison.
The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →#1 Best Overall
The “take on OpenAI” framing captures the competitive ambition behind the release, not proof that Llama 3.1 outperformed OpenAI’s models in every use case. The evidence here does not establish present-day benchmark standings, comparative latency, or which model is best for a particular application.
Can you run Llama 3.1 locally?
Meta made the model family available through its download site and Hugging Face, so the weights were released for people and organizations to obtain rather than being limited to a Meta-hosted interface. Whether you can run a particular variant locally depends on your hardware, software setup, quantization choices and workload. The release details establish parameter counts and context length, but not a minimum hardware configuration or performance target for a given computer.
When choosing a variant, weigh likely capability needs against available compute and acceptable response time. Also check whether your application needs the full 128K-token context; a maximum context window is a capability, not a guarantee that every deployment will handle that length at the same speed or cost.
Is Llama 3.1 really open source?
Meta presented Llama 3.1 as open source, but for licensing precision it is better to call it openly available or open-weight. Meta distributes it under a custom Llama 3.1 Community License rather than an unrestricted open-source license. The license grants limited, non-exclusive, worldwide, royalty-free rights to use, reproduce, distribute, copy, modify and create derivative works, subject to its terms. Read the Llama 3.1 Community License before using or redistributing the materials.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Repair Windows errors before they cause bigger problems3Scan for outdated or missing drivers - takes under a minuteRank #3
- Incredibly Light. Surprisingly Thin. - LG gram is designed to go wherever you do. Weighing just 2.5 lbs. with an ultra-slim 0.7-inch profile, it slips easily into your bag and feels light in hand—making it effortless to carry, commute, and work from anywhere.
- Remarkably Light. Reliably Strong. - LG gram has passed seven military-grade durability tests, striking an impressive balance between a highly portable, lightweight metal build and the confidence to handle everyday movement and travel.
- Power That Last with Smart Efficiency - LG gram combines a high-capacity 72Wh battery with AI-driven power management to optimize efficiency based on your usage. The result is up to 32 hours of video playback for} long-lasting performance that keeps up with your day—at home, at work, or wherever you go.
- AMD Ryzen AI Performance - Powered by AMD’s AI-optimized Ryzen processor with Radeon Graphics and a built-in NPU, LG gram delivers smooth multitasking and responsive performance. Fast 32GB LPDDR5x memory and 1TB NVMe storage keep everything moving without slowdowns.
- Dual AI for Always-On Intelligence - LG gram’s Dual AI—powered by EXAONE 3.5, LG’s AI solution—combines gram chat On-Device AI and gram chat Cloud AI to deliver seamless assistance. gram chat On-Device AI enables fast document search and summarization directly on your PC, while gram chat Cloud AI expands capabilities when connected—so everyday tasks stay smooth, responsive, and uninterrupted.
Conditions to understand before redistribution
- The agreement requires redistribution to include a copy of the license.
- It requires a prominent “Built with Llama” notice in an associated location and the specified attribution notice in a Notice file.
- At the July 23, 2024 release date, a licensee whose relevant products or services had more than 700 million monthly active users in the preceding calendar month was required to request a separate license from Meta.
These are a summary of selected terms, not legal advice. The license text controls, and the clause above is specifically the release-date condition described in the agreement; check the current agreement for any revisions before relying on it.
What should developers know about safety and deployment?
Meta’s model card says Llama models are not designed to be deployed in isolation and should be part of an AI system with additional safety guardrails as required. Developers are responsible for the safety of the systems they build, including integrations with tools. Meta identified Llama Guard 3, Prompt Guard and Code Shield as available safeguards in the release materials. These tools do not remove the need to assess risks in the full application, its inputs and its actions.
What did training the 405B model involve?
Meta reported that training across the Llama 3.1 family used 39.3 million H100-80GB GPU hours. For the 405B model, Meta reported estimated training emissions of 11,390 tons of CO2-equivalent under location-based accounting and 0 tons under market-based accounting. These are Meta’s training estimates and accounting figures; they are not the operating footprint of a deployed model or a user’s inference workload. Meta’s announcement provides the company’s stated figures and methodology context.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Where was Llama 3.1 available at launch?
Meta said the models were available through its download site and Hugging Face, and named more than 25 ecosystem partners, including AWS, NVIDIA, Databricks, Groq, Dell, Azure, Google Cloud and Snowflake. This describes Meta’s launch announcement; it does not establish present-day access, provider pricing, or the current availability of a particular hosted service. Check the provider’s own current documentation for those details. Meta’s launch announcement lists the ecosystem partners.
Quick Recap
Best Value
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
