Timeline
Reported development of free 'Spark' LLM targeting 1B users.
Meta Superintelligence Labs launched Muse Spark, a natively multimodal reasoning model, after a nine-month infrastructure rebuild.
Leaked information suggests Meta's upcoming 'Spark' model will be a closed-weight release, breaking from its open-source strategy.
Suspended collaboration with data-labeling vendor Mercor due to a supply chain security breach.
Publicly detailed the development and architecture of its Adaptive Ranking Model (ARM) for ads infrastructure.
Published research paper on Query-only Test-Time Training (QTT) method to fix 'lost-in-the-middle' problem in long-context LLMs
Launched unified coding subscription plan offering four AI models for $3
Qwen family captured over 50% of global open-source AI model downloads, reaching nearly 1 billion cumulative downloads on Hugging Face.
Released Qwen 3.5 flagship model series, positioned as on par with leading US proprietary models.
Released Qwen 2.5, marking the beginning of a decisive shift away from popular US models like Llama.