All tags
Company: "arcprize"
not much happened today
kimi-k3 claude-fable-5 opus-4.8 gpt-5.6-terra gpt-5.5 inkling glm-5.2 gpt-5.6-sol moonshot openai thinking-machines artificial-analysis arena datacurve arcprize aisecurityinst moe-routing quantization data-curation infrastructure-design coding-agents benchmarking front-end-development software-engineering arc-benchmarks cybersecurity zhilin_yang kimmonismus anikasomaia dylan522p novasarc01 scaling01 theo hqmank
Moonshot's Kimi K3 release has sparked a reassessment of Chinese open-weight models' proximity to the frontier, with strong performance in coding, agentic tasks, and long-horizon knowledge work. The strategic focus has shifted from a "compute moat" to an "efficiency stack" involving MoE routing, quantization, data curation, and scarcity-driven infrastructure like Moonshot's "Mooncake" stack. Benchmarks from Artificial Analysis, Arena, DeepSWE, ARC, and Cyber place K3 among the top models, with scores such as 57 on the Intelligence Index and coding agent benchmarks matching or surpassing models like GPT-5.6 Terra and Claude Fable 5. Discussions continue on K3's exact standing, but it is now widely recognized as a significant frontier contender.
not much happened today
arc-agi-3 claude-code anthropic langchain arcprize primeintellect agentic-reasoning interactive-environments benchmarking efficiency-metrics zero-preparation-generalization agent-infrastructure trainable-agents classifier-approval fchollet mikeknoop scaling01 _rockt mark_k andykonwinski bradenjhancock jeremyphoward togelius bracesproul hwchase17 caspar_br _catwu
ARC-AGI-3 benchmark introduced by @arcprize and François Chollet resets the frontier for general agentic reasoning with humans solving 100% of tasks versus under 1% for current models, focusing on zero-preparation generalization and human-like learning efficiency. The scoring protocol sparked debate over its harsh efficiency-based metric compared to prior ARC versions and other benchmarks like NetHack. The community acknowledges the benchmark highlights weaknesses in current LLM agents in interactive, sparse-feedback environments. Concurrently, agent infrastructure advances with LangChain launching Fleet shareable skills for reusable domain knowledge, and Anthropic revealing Claude Code auto mode for classifier-mediated approval balancing autonomy and manual confirmation. Browser and coding agents are evolving into trainable systems beyond prompt wrappers, exemplified by BrowserBase and Prime Intellect collaboration.
new Gemini 3 Deep Think, Anthropic $30B @ $380B, GPT-5.3-Codex Spark, MiniMax M2.5
gemini-3-deep-think-v2 arc-agi-2 google-deepmind google geminiapp arcprize benchmarking reasoning test-time-adaptation fluid-intelligence scientific-computing engineering-workflows 3d-modeling cost-analysis demishassabis sundarpichai fchollet jeffdean oriolvinyalsml tulseedoshi
Google DeepMind is rolling out the upgraded Gemini 3 Deep Think V2 reasoning mode to Google AI Ultra subscribers and opening early access to the Vertex AI / Gemini API for select users. Key benchmark achievements include ARC-AGI-2 at 84.6%, Humanity’s Last Exam (HLE) at 48.4% without tools, and a Codeforces Elo of 3455, showcasing Olympiad-level performance in physics and chemistry. The mode emphasizes practical scientific and engineering applications such as error detection in math papers, physical system modeling, semiconductor optimization, and a sketch to CAD/STL pipeline for 3D printing. ARC benchmark creator François Chollet highlights the benchmark's role in advancing test-time adaptation and fluid intelligence, projecting human-AI parity around 2030. This rollout is framed as a productized, compute-heavy test-time mode rather than a lab demo, with cost disclosures for ARC tasks provided.