Startup Subquadratic Claims Breakthrough in LLM Efficiency
Miami-based startup Subquadratic claims its new model SubQ is 56 times faster than FlashAttention models in speed tests, with a context window up to 12 million tokens.
Model Release
214 articles
New AI models as they ship — large language models, multimodal systems, and open-weight releases. Capabilities, context windows, pricing, and availability, summarized so you can compare what is actually new.
LATEST NEWS
-
-
OpenAI's GPT-5.5 Instant Outperforms Doctor-Written Health Answers
OpenAI's GPT-5.5 Instant model outperforms doctor-written health answers by 71% in accuracy, clarity, and completeness, according to the company.
-
OpenAI Launches GPT-5.5 Instant for Enhanced Health Intelligence
OpenAI's GPT-5.5 Instant improves health-related responses, matching frontier models' performance in critical evaluations, with a 71% drop in factuality issues over two months.
-
Grok 4.3 Now Available on Amazon Bedrock
Grok 4.3, the latest version of the model, is now generally available on Amazon Bedrock, achieving the lowest hallucination rate among frontier models.
-
Grok Imagine Video 1.5 Now Generally Available
xAI has launched Grok Imagine Video 1.5, offering faster generation times and enhanced audio and motion quality for creators.
-
Anthropic Halts Mythos 5 Amid U.S. Export Controls
Anthropic paused its Mythos 5 and Claude Fable 5 models last week after U.S. export controls blocked foreign access. The company is seeking White House approval to resume service.
-
Amazon Bedrock Now Offers Gemma 4 Models
Amazon Bedrock has launched Gemma 4 models, including a 31B variant with an Intelligence Index of 39, surpassing the 15 median in the 4B–40B class.
-
Loft Orbital Satellite Uses Gemma 3 VLM to Find Objects on Its Own
A satellite launched by Loft Orbital in 2025 used Google DeepMind’s Gemma 3 vision-language model to identify objects autonomously in April, marking the first use of such technology in orbit.
-
Microsoft's Mirage Model Enhances Video Generation with Spatial Memory
Microsoft Research's Mirage model generates videos up to 10.5x faster and uses 55x less memory than comparable systems, according to thedecoder.
-
Microsoft Announces Azure AI Model Enhancements at Build 2026
Microsoft unveiled new AI model updates at Build 2026, including improved efficiency and integration with Copilot.
-
Tsinghua University Unveils 'Count Anything' AI Model
Tsinghua University researchers introduced 'Count Anything,' an AI model that counts and labels objects across diverse image types with high accuracy, outperforming competitors in benchmark tests.
-
Microsoft CEO Nadella Admits to Token-Maxing Addiction
Microsoft CEO Satya Nadella warns against token-maxing but admits to being addicted to using large AI models for various tasks.