Model launchesNew models: DeepSeek V4.1 Flash 1017, MiniMax M3.1 Flash and Mercury 3 released, limited adoption expectedThe platform Latent Space reported in a paid post the arrival of new models: DeepSeek V4.1 Flash 1017, MiniMax M3.1 Flash and Mercury 3.1 min read
Model launchesAnthropic launches Claude Gov models for U.S. national security customersAnthropic announced on June 6, 2025 that it has deployed a custom set of Claude Gov models exclusively for U.S.2 min read
Model launchesNew models markedly improve coding and migration performance for low‑resource programming languagesRecent model updates substantially raise accuracy and lower cost for working with low-resource languages such as COBOL, Fortran, OCaml and Visual Basic.4 min read
Model launchesNvidia unveils open-weight Nemotron Super VL and Bolt models for enterprise agents and local AI PCsNvidia introduced new open-weight models aimed at always-on enterprise agents and local AI PCs.3 min read
Model launchesMeta's Muse Spark Open Narrows US–China Open-Model GapMeta released Muse Spark Open, an open-weights model that significantly strengthens American offerings in the open-model space.4 min read
Model launchesGPT-6 Astra Pro receives update in ChatGPT for faster and parallel agent workOpenAI updated the GPT-6 Astra Pro model in the ChatGPT service: it now provides faster responses, uses plugins and tools more efficiently, and allows multiple agents to work on a task in parallel.1 min read
Model launchesReflection unveiled Beam, a U.S.-made open-weight AI model, now free on the Vercel AI GatewayBeam, an American-made open-weight AI model developed by Reflection, is now available for free on the Vercel AI Gateway (2026).1 min read
Model launchesApple and Meta to Release an Open-Weight AI Model for High-End MacsApple and Meta announced a joint effort to produce an open-weight artificial intelligence model designed to run on high-end Mac computers.2 min read
Model launchesElorian releases Glimpse, a model that repeatedly returns to images during visual reasoningElorian, a Palo Alto AI startup that previously raised $55 million in seed funding at a $300 million valuation, has published a research-preview model called Glimpse.3 min read
Model launchesVercel: Jev becomes fastest-adopted decision-focused AI modelVercel’s usage data shows that Jev, a decision-oriented model from TypeSafe AI released on September 15, reached 35% of paid teams within two weeks — nearly twice the reach of Sonnet 5.5.3 min read
Model launchesReflection releases Beam — a 501B-parameter MoE model with 1M-token context and open-source toolingReflection published Beam on Hugging Face: a mixture-of-experts (MoE) language model with 501 billion parameters (23B active per token), a 1,000,000-token context window, and Apache 2.0 licensing.5 min read
Model launchesMeta Superintelligence Labs' Muse Wonder now available in CursorMeta Superintelligence Labs' Muse Wonder model is now available on the Cursor interface; on the CursorBench baseline effort it scored 46.1%, the best performance among the Muse models.1 min read
