#Ai2

SERA Agents Slash Cost of Private Codebase AI Specialization

Ai2’s SERA models use Soft-verified generation to dramatically reduce the cost of specializing AI agents to private codebases, making enterprise deployment accessible.

HiRO-ACE Delivers Accessible Kilometer-Scale AI Climate Simulation

The HiRO-ACE AI climate simulation framework provides decades of 3km resolution regional climate data in a single day, solving the major computational barrier for high-fidelity modeling.

Bolmo Advances Byte-Level Language Models with Practicality

NeuroDiscoveryBench Sets New Standard for Neuroscience AI Benchmarks

Olmo 3 Open-Source AI Unlocks Full Model Flow Transparency

DR Tulu deep research: Open AI closes proprietary gap

OlmoEarth Redefines Earth Observation Foundation Models

Ai2’s new AI climate emulator runs 1,500 years in a day

AI Climate Emulator Slashes Energy Consumption 3,750x

NVIDIA, NSF Partner to Fuel Open AI for US Scientific Leadership

NVIDIA is collaborating with the U.S. National Science Foundation (NSF) to bolster scientific research through a new AI initiative. It aligns directly with the White House AI Action Plan, particularly the "Winning the AI Race: America’s AI Action Plan"

Ai2 Debuts Top Open Source Foundation Model OLMo 2 with Open Weights, Data, and Code

<p>OLMo 2 delivers breakthrough performance across benchmarks.</p><p>ARC Challenge (commonsense reasoning): OLMo-2-13B scores 83.5, outperforming Llama-3.1-8B (79.5) and Qwen-2.5-7B (67.4).</p><p>MMLU Massive Multitask Language Understanding (domain-specific knowledge): OLMo-2-13B achieves 67.5, higher than Qwen-2.5-7B (64.4) and Llama-3.1-8B (66.9).</p><p>GSM8k Math Word Problems (mathematical reasoning): OLMo-2-13B scores 75.1, significantly outperforming Llama-3.1-8B (51.3) and Qwen-2.5-7B (63).</p><p>TriviaQA (knowledge recall): OLMo-2-13B achieves 81.9, comparable to Qwen-2.5-7B (81.5) and higher than Llama-3.1-8B (80.3).</p>