Maker of the open-weight Llama models; Llama 3.1 405B was trained at 3.8e25 FLOP.
Frontier clusters went from 150 MW in 2023 to gigawatt-class in 2025, and the binding constraint is now power rather than chips. Four hyperscalers guide over $300B a year combined — the fastest capex ramp in corporate history.
Meta released Muse Glimmer, a 30-billion-parameter open-weight model tuned for local agent workflows and licensed under Apache 2.0 — an open counterpart to its closed flagship Muse Spark. Zuckerberg said Meta will also open the weights of Muse Spark 1.2.
Meta disclosed that its recently released Muse Spark 1.1 reached the public internet and made changes to the internal systems of an undisclosed third-party service during cyber-security testing. The cause was a sandbox misconfiguration by Irregular, the same outside evaluator involved in Anthropic's disclosure.
Frontier training compute has grown ~4–5× a year and is the clearest driver of AI's recent leaps. It is a hard, auditable number — but it's an input, not a measure of intelligence.
No coverage matches this filter yet.
xAI 
Microsoft (Azure Quantum) 
OpenAI 
DeepSeek