TestingCatalog News 🗞️ Telegram Posts

MiniMax unveils MiniMax-01 Series 2 with 456B-parameter AI model

MiniMax has launched the MiniMax-01 Series 2, featuring the MiniMax-Text-01, a 456B parameter MoE model with advanced attention mechanisms and long-context capabilities. Open-sourced, it competes strongly in long-context tasks, though refinement in coding remains needed.

🗞 #ai

OpenAI introduces Tasks beta for ChatGPT, enabling scheduled actions

OpenAI's new ChatGPT Tasks beta lets Plus, Pro, and Teams users schedule one-time or recurring activities like reminders or updates. Notifications are sent across platforms, with management limited to the web interface and capped at 10 active tasks during beta.

🗞 #chatgpt

New Codestral 25.01 sets FIM performance benchmark in AI coding

Mistral AI's Codestral 25.01 offers faster, more accurate coding capabilities with improved architecture, 256k token context length, and top benchmark performance. Accessible via IDE plugins, APIs, and enterprise platforms, it aims to optimize modern software development.

🗞 #mistral

DeepSeek launches iOS app, challenging ChatGPT with free AI tools

DeepSeek has released a free iOS app combining its DeepThink reasoning AI and AI-powered search tool. Powered by the DeepSeek V3 model, it positions the company as a competitive alternative to ChatGPT, with plans for broader platform availability.

🗞 #deepseek

Perplexity rolled out real-time updates for NBA and NFL fans

Perplexity Sports introduced real-time updates, game schedules, play-by-play analysis, and sports news for NBA and NFL, aiming to centralize content for fans. Announced during NFL playoffs, it reflects Perplexity AI's shift toward niche AI-driven sports analytics.

🗞 #perplexity

Mind mapping feature spotted in Google's NotebookLM under tests

Google is testing an AI-driven mind map feature for NotebookLM to visually organize notebook sources. It supports JSON export for external tools and aligns with its purpose of managing complex information, though its release timeline remains uncertain.

🗞 #notebooklm

OpenAI tests shareable Canvas and revamped Custom Instructions

OpenAI is updating ChatGPT with a Canvas sharing option for collaboration, a redesigned Custom Instructions UI for tailored responses, and a model selector on iOS for switching GPT models. These changes aim to broaden usability and align tools across platforms.

🗞 #chatgpt

Luma AI tests Ray-2 model with 1080p video support

Luma AI's "Ray-2" text-to-video model offers 1080p resolution and 20-second outputs, an upgrade from Ray-1. Currently accessible to select creators, its broader release could reshape video creation in marketing, entertainment, and social media. Launch timing remains unconfirmed.

🗞 #ai

Qwen Chat launch: Free AI tools for coding, reasoning, and more

Alibaba has launched Qwen Chat, a web platform for interacting with its Qwen AI models. Features include multimodel comparisons, document and image analysis, HTML previews, and coding support. Future updates may add web search, image generation, and voice functionality.

🗞 #ai

OpenAI explores similar video search for Sora users

OpenAI is testing a video search feature for its AI tool, Sora, enabling users to find similar videos in their feed. Still in development, the feature aims to refine content discovery and complement Sora’s text-to-video and storyboard tools.

🗞 #sora