Google DeepMind unveiled a new research paper demonstrating significant advances in AI reasoning and planning, with a model capable of solving complex logic puzzles and real-world planning tasks with higher accuracy than previous systems.
GPT-5.2 is rolling out to all ChatGPT users starting with paid plans, offering 3,000 messages per week with GPT-5 Thinking mode for Plus users.
Claude Code received updates with sub-agents that delegate tasks to specialized AI assistants, and Anthropic plans further investment after reaching $1B run rate in six months.
Google's Gemini 3 Pro unveiled with a historic 1501 Elo score on LMArena, the highest ranking ever, and Gemini 2.5 Flash achieves 372 tokens/second for fast applications.
Anthropic's Claude Opus 4.5, launched in November 2025, achieved a record 80.9% on SWE-bench Verified, leading in coding performance benchmarks.
OpenAI released GPT-5.2 on December 11, 2025, introducing three operating modes: Instant for low latency tasks, Thinking for deep reasoning, and Pro for maximum intelligence, with a 400K token context window and memory feature across sessions.
This new technique enables LLMs to dynamically adjust the amount of computation they use for reasoning, based on the difficulty of the question.[3]
CSAIL researchers find even “untrainable” neural nets can learn effectively when guided by another network’s built-in biases using their guidance method.[3]
MIT-IBM Watson AI Lab researchers developed an expressive architecture that provides better state tracking and sequential reasoning in LLMs over long texts.[3]
Google introduced Nano Banana for precise image editing in Gemini by drawing, circling, or annotating directly on images, alongside NotebookLM integration and expansions to Wear OS, Google TV, and Android Auto.
Anthropic updated Claude Code with sub-agents that delegate tasks to specialized AI assistants for independent contexts.