SID AI has released SID-2, a search model that replaces fixed RAG pipelines with RL-trained agents capable of searching, reading and re-searching. Its Swarm variant forks context across multiple small agents that message each other to explore different corpus regions. On SID's 506-question benchmark, a 6-agent Swarm reaches 0.701 NDCG in 17.7s versus GPT-6 Astra's 0.704 in 135s, while the 12-agent version costs $515 per 1,000 queries, nearly 4x Opus 5.5. The model is available via API and Baseten, but parameter counts and weights are undisclosed and all scores are self-reported.
Agent Swarm
智能体集群
Multi-agent architecture where small agents fork context, message each other and explore corpus regions in parallel.
Curated Intelligence
Google has opened Gemini 4 Carbon to the general public just two weeks after a limited release to cybersecurity defenders, calling it its strongest model yet and positioning it for complex multi-step coding and open-ended scientific research. It also previewed a swarm feature on its Antigravity agent development platform, complementing last month's Argon release.
Hugging Face chief policy officer Irene Solaiman warns that the openness and rich data that make open AI platforms valuable also make them more vulnerable to attacks. Recent incidents, including the October breach of OpenCode that targeted users' third-party API keys and caused financial damage, and a summer incident in which an OpenAI agent swarm breached Hugging Face, underscore the risk. Security experts recommend prioritizing detection of suspicious behavior and containment capabilities over relying solely on authentication.
Axiom Math utilized an agent swarm to formalize the entire 'Stacks Project' into 16 million lines of Lean code, creating a computer-verifiable foundation for modern algebraic geometry. By orchestrating agents like an engineering team, they ensured rigorous logical validation and open-source accessibility, providing critical infrastructure for AI-assisted mathematical proofs.
Researchers from Y Combinator and Olam Labs tested hundreds of AI agents in Minecraft for collaboration and combat. Results showed that multi-agent collaboration often devolves into inefficient 'bureaucracy,' while GPT-6.1 Sol swarms defeated Claude Opus 5.5 clusters in PvP due to aggressive, hive-mind strategies. The open-source study aims to reveal behavioral patterns of agent swarms relevant to real-world applications.
Researchers from Y Combinator Labs and Olam Labs deployed hundreds of AI agents in Minecraft to test collaboration and adversarial scenarios. The study revealed that while large-scale agent swarms excel at solving mathematical problems, they fail to build complex civilizations due to excessive coordination overhead and 'bureaucracy.' These experiments aim to provide insights into the bottlenecks of multi-agent systems in real-world applications.
Axiom announced that a swarm of AI agents completed the Lean formalization of the Classification of Finite Simple Groups (CFSG) in just 24 days, generating approximately 40 million lines of code. This milestone marks verifiable AI's transition from lab experiments to scalable applications, demonstrating the immense potential of multi-agent collaboration in complex logical reasoning tasks.
At CPE 2026, Damao Tech unveiled its full-stack compute-power synergy system featuring vertical energy LLMs and agent clusters. The company proposes an E-AIDC paradigm to resolve green power direct connection bottlenecks via multi-market arbitrage, aiming to optimize the energy efficiency and carbon performance of AI data centers.
Security firm Norforge revealed that a hacker deployed an autonomous swarm of AI agents to breach over 100 corporate websites, including John Deere and SharkNinja, with minimal human intervention. This incident highlights the operationalized risk of highly autonomous AI agents in cyberattacks, marking a new phase for automated threats.
Security firm Norforge revealed that a hacker deployed an autonomous swarm of AI agents to breach over 100 websites, including John Deere and SharkNinja, with minimal human intervention. This incident highlights the operational capability of AI agents in automated cyberattacks and raises new concerns about AI safety boundaries.