AWS Unveils Reflection Beam Model for Agentic Coding
AWS released Reflection Beam, an open-weight model for agentic coding, with 501 billion total parameters, available in US East regions.
735 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.
AWS released Reflection Beam, an open-weight model for agentic coding, with 501 billion total parameters, available in US East regions.
Mistral and Mozilla announced Firefox Smart Window powered by Mistral models, available in France and North America starting September 2026.
HuggingFace released autotrust/JEV-27B-VL, a model that generates calibrated probabilities for short-video recommendations, outperforming traditional systems by 0.010 on 1,000 decisions.
Hugging Face released Apertus 1.5 on July 24, 2026, expanding the 8B and 70B models with image and audio understanding, a 4x longer context window, and improved instruction-following.
WIRED reporter found Meta's AI agent Muse capable of sexting, with millions of users downloading the app in recent weeks.
OpenAI announced the Rosalind Academic Plan on October 15, 2026, offering eligible labs free access to GPT-Rosalind for up to 12 months.
Nvidia released Nemotron 3.5 Super VL, an open multimodal model with 120 billion parameters, designed for continuous enterprise AI agent operations.
Black Forest Labs released FLUX 3 Image Klein, a 5.6-billion-parameter model that runs on consumer GPUs with 8 GB of VRAM, achieving high Elo ratings.
Meta released Muse Spark Open, an open-weight model with 816 billion parameters, on October 15, 2026, claiming it outperforms Qwen 3.8, GLM 5.3, and Kimi K3 in several categories.
Meta’s Muse Spark Open, a coding-optimized model with 816 billion parameters, is now available in Amazon SageMaker JumpStart, offering developers a 1 million token context window.
Mistral released Timestral on October 15, 2026, with its largest model achieving top rankings on three key benchmarks using just 209M parameters.
Artemis Security unveiled Orion-1, an AI model that outperforms general-purpose models like GPT-6 Sol on its own benchmark with an 80.8 score in attack reconstruction.