Suno has introduced its new AI music model v5.5 along with features designed to showcase musical individuality
By Chris Borden Google DeepMind is acquiring a minority stake in the studio behind the space MMO EVE Online and plans to use the game to test AI models. At the same time, developer CCP Games is buying itself out from its South Korean owner Pearl Abyss for $120 million — less than the $225 million Pearl Abyss paid for the studio in 2018 — and rebranding as Fenris Creations.
By Chris Borden Google has released Multi-Token Prediction Drafters (MTP) for its open-source Gemma 4 model family, designed to accelerate text generation by up to three times.
By Daniel Mercer OpenAI is replacing ChatGPT's current default model with GPT-5.5 Instant. The update is designed to be more factually accurate, more concise, and more personalized. New Memory Sources let users see for the first time exactly what context is influencing their answers.
By Chris Borden With MiMo-V2.5-Pro, Xiaomi has released an AI model that — according to internal tests — writes a complete compiler in under five hours and rivals Anthropic's Claude Opus 4.6 on coding benchmarks, while consuming significantly fewer tokens than its Western competitors.
By Daniel Mercer OpenAI investigated a strange behavior in its AI models: starting with GPT-5.1, the models increasingly began using goblins, gremlins, and other mythical creatures in their responses. Mentions of "goblin" rose by 175% following the launch of GPT-5.1.
By Alex Rowland Chinese AI startup DeepSeek has published a preview of its new family of language models. The flagship model, V4-Pro, is presented as a major open-weight AI system that reportedly outperforms Claude Opus 4.6 and GPT-5.4 in several key benchmarks.
By Alex Rowland OpenAI has released GPT-5.5, positioning the model as a new level of intelligence for real-world work, complex task execution, and AI agent management.
By Chris Borden Anthropic’s Opus 4.7 carries the same list pricing as Opus 4.6 on paper. In practice, however, it appears to consume noticeably more tokens per request. That is according to measurements published by developer Abhishek Ray on Claude Code Camp.
By Chris Borden Moonshot AI has released Kimi K2.6 as an open-weight model. According to the company, it can compete with GPT-5.4, Claude Opus 4.6, and Gemini 3.1 Pro on several coding and agent benchmarks. Reported scores include 54.0 on HLE with Tools, 58.6 on SWE-Bench Pro, and 83.2 on BrowseComp. Moonshot also says K2.6 can execute more than 4,000 tool calls and run continuously for over 12 hours on long-horizon tasks in languages such as Rust, Go, and Python.
ES 
EN