Model launchesTencent unveils Hy3 MoE large language model with 256K context lengthTencent's Hy Team has released Hy3, a Mixture-of-Experts large language model with 295 billion parameters (21B active) and a 256K token context window.2 min read
Model launchesRacka: a Hungarian-optimized, reasoning-capable experimental LLMRacka is a Hungarian-language large language model (Racka-4B) developed by researchers at Eötvös Loránd University’s Department of Artificial Intelligence with the Digitális Örökség Nemzeti Laboratórium and in collaboration with Mynds.ai.5 min read
Model launchesAnthropic introduces Claude 3.7 Sonnet with optional “extended thinking” and visible thought tracesAnthropic has released Claude 3.7 Sonnet (announced Feb 24, 2025), which adds an opt‑in “extended thinking mode” that lets the model spend more computation and time on difficult problems, and optionally shows its intermediate thought process.6 min read
Model launchesMicrosoft unveils MAI-Thinking-1, an in-house reasoning language modelMicrosoft introduced MAI-Thinking-1, its first reasoning language model developed independently rather than distilled or fine-tuned from another developer’s model.4 min read
Model launchesAnthropic launches Claude Fable 5 and restricted Claude Mythos 5 with strict safeguards and limited accessAnthropic released two Mythos-class models: Claude Fable 5, widely available but guarded by conservative safety classifiers that fallback to Claude Opus 4.8 for sensitive topics, and Claude Mythos 5, the same core model with some safeguards relaxed for vetted cyberdefense and biomedical partners.5 min read
Model launchesClaude Fable 5 available again on Cursor, leads on CursorBench but most expensive per taskThe Claude Fable 5 model is available again on the Cursor platform; it currently leads in performance on CursorBench, but it incurs the highest cost per task.1 min read
Model launchesFor Claude fans: Sonnet 5 has arrived, Fable returns tomorrowAnthropic announced the release of Sonnet 5 for Claude users and said that the Fable model will return tomorrow; according to the statement there was a lot of misunderstanding around Sonnet 5, so they also tried to clarify the details in a video.1 min read
Model launchesGoogle’s Gemini 3.1 Flash Lite Image (Nano Banana 2 Lite) released as a fast, low-cost image modelGoogle’s Gemini 3.1 Flash Lite Image, marketed as Nano Banana 2 Lite, is presented as the fastest and cheapest Gemini image model optimized for speed and scale.2 min read
Model launchesAnthropic releases Claude Sonnet 5 with 1M-token context and new tokenizer that raises effective costsAnthropic published Claude Sonnet 5 with a million-token context window, up to 128,000 output tokens, and several API changes.3 min read
Model launchesClaude Sonnet 5 available on Cursor, better results on CursorBenchThe Claude Sonnet 5 model is now available on the Cursor platform.1 min read
Model launchesTabFM: a foundation model for tabular prediction using in‑context learningTabFM reframes tabular classification and regression as an in‑context learning (ICL) problem, allowing one‑pass predictions on previously unseen tables without per‑dataset training, hyperparameter tuning, or manual feature engineering.4 min read
Model launchesMeituan releases 1.6-trillion-parameter LongCat-2.0 with 1M-token context and China-made ASIC trainingMeituan has published LongCat-2.0, a 1.6-trillion-parameter Mixture-of-Experts model with a native one-million-token context window, under an MIT license and commercially available pricing.6 min read
