Taalas embeds the Llama 3.1 8B model directly into a 6nm ASIC, delivering 17,000 tokens per second—nearly ten times faster than top NVIDIA GPUs—while cutting system cost by over tenfold and power consumption by tenfold, albeit with limited flexibility and quantization trade‑offs.
A UC San Diego study found that GPT-4.5 was judged human 73% of the time and LLaMa-3.1-405B 56%, demonstrating that both large language models can pass a standard three‑party Turing test, with detailed methodology, results, and analysis of judge behavior.
A leaked 405‑billion‑parameter Llama 3.1 model shows mixed benchmark results—outperforming GPT‑4o on some tasks while lagging on others—along with massive hardware requirements, extensive training data, and new safety considerations that could reshape AI deployment.
Meta has officially released Llama 3.1, a 405‑billion‑parameter open‑source model that matches or surpasses GPT‑4o and Claude 3.5 on over 150 benchmarks, expands context to 128 K tokens, supports eight languages, and is accompanied by a detailed 100‑page paper describing its data, training stack, architecture, quantization, safety measures, and ecosystem support.
Meta’s latest open‑source Llama 3.1 model, available in 8B, 70B, and 405B sizes, is evaluated against top competitors and can be easily run locally on the 8B version using Ollama with a simple step‑by‑step guide.
Meta’s newly released Llama 3.1 series, highlighted by the 405B model trained on 150 trillion tokens, claims state‑of‑the‑art performance in coding, mathematics, and multilingual summarization while offering an open‑source alternative to GPT‑4o and Claude 3.5 Sonnet.
