Nvidia's Nemotron-Cascade 2 wins math and coding gold medals with 3B active parameters — and its post-training recipe is now open-source

Nvidia's Nemotron-Cascade 2 is a 30B MoE model that activates only 3B parameters at inference time, yet achieved gold medal-level performance at the 2025 IMO, IOI, and ICPC World Finals. Nvidia has ...
SiliconANGLE

Google revamps Gemini 2.5 Pro again, claiming superiority in coding and math

Google LLC has launched another, even more capable preview of its powerful Gemini 2.5 Pro model, proclaiming it to be the “most intelligent” large language model it has released so far. Today’s is the ...
Android Authority

The latest Gemini model is a coding and math whizz with 'parallel thinking'

Google has announced Gemini 2.5 Pro Deep Think. This AI model is said to offer “incredible” performance in advanced math and coding tasks. The new AI model will be available to Gemini AI Ultra ...

Meta researchers introduce 'hyperagents' to unlock self-improving AI for non-coding tasks

Meta's new hyperagent framework breaks the AI "maintenance wall," allowing systems to autonomously rewrite their own logic ...
EdSurge

Teaching Machines to Spot Human Errors in Math Assignments

When completing math problems, students often have to show their work. It’s a method teachers use to catch errors in thinking, to make sure students are grasping mathematical concepts correctly. New ...
Digi Times

Alibaba upgrades Qwen3 series, surpassing OpenAI and DeepSeek in math and coding capabilities

China's open-source artificial intelligence sector has made significant new strides. Alibaba has updated its Qwen3 series of large language models, outperforming OpenAI GPT-4o, DeepSeek V3, and ...