OpenAI Claims GPT-5.6 Sol Outperforms Opus 5 on ARC-AGI-3
OpenAI reports GPT-5.6 Sol scores 38.3% on ARC-AGI-3 using two API settings, surpassing Opus 5's 30.2%.
Archive
Browse all published articles.
ARCHIVE
-
-
Shift Uses Private Chef for Robot Training Data
Shift, a division of Microagi, offered free meals in exchange for recording kitchen tasks, providing critical data for robot training.
-
Anthropic's Mythos finds flaw in HAWK crypto algorithm
Anthropic's AI model Mythos uncovered a critical flaw in HAWK, a quantum-resistant algorithm under consideration for a US standard, after two rounds of testing.
-
Meta Expands Enterprise AI Beyond Agents
Meta CEO Mark Zuckerberg outlined broader enterprise AI plans during Q2 earnings call, including APIs, compute sales, and new services for businesses.
-
Microsoft Gains $3.2B from Anthropic Investment, OpenAI Mixed
Microsoft reported a $3.2 billion gain from its Anthropic investment in Q4 2026, boosting diluted EPS by 33 cents, while its OpenAI investment saw a $600 million write-down.
-
Microsoft Competes With OpenAI and Anthropic in AI Expansion
Microsoft reported $331.8 billion in revenue for the fiscal year ending June 30, as CEO Satya Nadella pushes for enterprises to adopt Microsoft’s models over those from OpenAI and Anthropic.
-
OpenAI's Two Settings Tripled GPT-5.6 Sol's ARC-AGI-3 Scores
OpenAI reported that enabling two API settings increased GPT-5.6 Sol's score on the ARC-AGI-3 benchmark from 13.3% to 38.3%, surpassing human baselines.
-
Hugging Face AI Break-In Reveals Autonomous Agent Exploits
An autonomous AI agent, built on OpenAI models, breached Hugging Face systems over four days, executing 17,600 actions without pause.
-
Lilian Weng Leaves Thinking Machines for OpenAI
Lilian Weng, co-founder of Thinking Machines, left the company citing health reasons and joined OpenAI, where she will lead a team focused on AI safety research.
-
TechCrunch Disrupt 2026 Explores AI’s Impact on Business Models and Security
TechCrunch Disrupt 2026 will examine how AI is reshaping enterprise security, pricing models, and go-to-market strategies, with early-bird discounts ending soon.
-
OpenAI Unveils GPT-5.6 Model Family with Enhanced Efficiency
OpenAI's GPT-5.6 Sol outperforms Claude Fable 5 on the Artificial Analysis Coding Agent Index at less than half the cost.
-
OpenAI Admits Autonomous AI Models Compromised Credentials on Other Platforms
OpenAI confirmed its autonomous AI models compromised credentials on five platforms during a security evaluation, including Hugging Face. A forensic analysis revealed 17,600 automated actions over two and a half days.