New AI models, benchmarks, open weights and frontier model updates curated by ChAIcked.
Polimill deployed OpenAI GPT models to help Japanese local governments search and apply administrative knowledge more efficiently. The initiative accelerates municipal digital transformation and knowledge management.
News · LLMs · Government · Infrastructure · — Open article
An examination of why Claude and similar LLMs produce inaccurate code time estimates, with strategies for better communication between developers and AI coding assistants.
Blog · LLMs · AI Coding · Development · — Open article
Guide on using Gemini Omni 1.1 Flash's scene extension feature to incrementally expand 10-second video clips up to 40 seconds total duration.
Blog · Models · Video · Development · — Open article
Jensen Huang announced Nvidia achieved AGI on an earnings call, then immediately called the milestone senseless. His comments highlight the lack of consensus on what AGI actually means in practice.
News · Models · Opinion · — Open article
Google DeepMind releases Gemini Omni 1.1 Flash with expanded control options for developers building with the model. The update focuses on improving customization and fine-tuning capabilities.
Blog · Models · Development · LLMs · — Open article
A developer shares decision-making frameworks for choosing between LLM-powered solutions and traditional deterministic methods when building AI-assisted features, using a support agent as a case study.
Blog · LLM · Application Design · Development · — Open article
A technical analysis examines which tokens form the essential vocabulary backbone of Anthropic's Claude model. The research reveals how language models distribute semantic weight across their tokenization space.
Blog · Models · Interpretability · Research · — Open article
Nvidia has agreed to acquire Hugging Face, a major open-source AI model hub and collaboration platform. The deal values the company at 13 billion dollars and signals consolidation in the AI infrastructure and model distribution space.
News · Acquisitions · Models · Infrastructure · — Open article
Google updated Gemini Audio with new 3.5 models offering improved transcription with automatic jargon detection, support for 85+ languages, and robust noise handling.
News · Models · Audio · Release · — Open article
Analysis of hidden costs and operational challenges in free-tier AI services, using MonkeyCode as a case study. Explores token consumption patterns, rate limits, and infrastructure trade-offs when offering free model access.
Blog · Cost Analysis · Infrastructure · Models · — Open article
IBM has unveiled Granite 4.2, a new generation of locally deployable large language models designed for enterprise use. The release prioritizes agentic capabilities and predictable performance in on-premises environments.
News · Models · Agentic AI · Enterprise Deployment · — Open article
Z.ai has confirmed that Ox Alpha is a new addition to its GLM model series and plans to release the model weights publicly. This move expands the availability of open-weight alternatives in the competitive large language model landscape.
News · Models · Open Source · — Open article
Researchers introduce AQLoRA, a method that accelerates quantized LoRA fine-tuning by using a single CPU pass to configure layer-wise quantization settings, eliminating the need for search or calibration data while reducing memory overhead.
Blog · Fine-Tuning · Quantization · LLMs · — Open article
PuzzleKV addresses memory bottlenecks in long-context LLM inference by applying page-wise low-rank decomposition to compress key-value caches, reducing storage requirements while maintaining model performance.
Blog · Inference Optimization · LLMs · Memory Efficiency · — Open article
A technical guide on training and optimizing multi-vector embedding models using the Sentence Transformers library. The tutorial covers practical approaches for building semantic search and similarity systems.
Blog · Development · Models · Fine-Tuning · — Open article
Loveholidays integrated OpenAI Codex to democratize software development across non-technical teams. The deployment accelerates product delivery by enabling broader organizational participation in coding tasks.
News · Code Generation · Developer Tools · LLMs · Enterprise · — Open article
A developer benchmarks the local Qwen 3.8 27B model across two agent shells, finding significant differences in output quality, token efficiency, and stability. PI Agent outperforms OpenCode without the 32k output ceiling.
Blog · Models · Agent Development · Benchmarks · — Open article
A developer created a local AI inference system for vehicle applications using a Raspberry Pi and Alibaba's Qwen language model. The project demonstrates edge deployment of LLMs on resource-constrained hardware.
News · Edge AI · LLMs · Hardware · — Open article
A Semi-Analysis newsletter report claims OpenAI's Jalapeño model achieves superior performance compared to Nvidia's Blackwell architecture. The claim carries implications for competitive positioning in AI hardware and model development.
News · Models · Hardware · Performance · — Open article
OpenAI unveiled its Jalapeño AI chip, claiming superior performance in balancing latency and throughput compared to competing inference hardware, addressing a traditional trade-off in AI system design.
News · Hardware · Inference · Models · — Open article