A developer detailed how they deployed self-hosted open-weight LLMs for a fintech client whose compliance team banned external AI APIs, running a 70B-class model on two GPU boxes for document Q&A, sup…
Microsoft has retired Phi-3.5-MoE-instruct on Azure AI Foundry, with the retirement schedule now listing only Phi-4 and no longer publishing an exact date. The model was delisted from models.dev's Azu…
Microsoft has retired the Phi-3-mini-instruct (4k) model from Azure AI Foundry, with the retirement schedule now listing only Phi-4 and no longer publishing an exact date. The model was delisted from …
Microsoft retired Phi-3-mini-instruct (128k), its phi-3-mini-128k-instruct model, from Azure AI Foundry hosted deployments, with the retirement schedule now listing only Phi-4 and no longer publishing…
Microsoft has retired Phi-3-small-instruct (8k), its 8k-context small language model, from Azure AI Foundry hosted deployments, with the retirement schedule now listing only Phi-4 and no longer publis…
Microsoft has retired Phi-3-small-instruct (128k), its azure-hosted deployment on Azure AI Foundry, as part of the Phi-3/3.5 MaaS retirement in 2025, with the model delisted from models.dev's Azure ca…
Microsoft has retired Phi-3-medium-instruct (4k), its 4k-context language model, from Azure AI Foundry hosted deployments as part of the Phi-3/3.5 MaaS retirement in 2025, and the model was delisted f…
Microsoft has retired Phi-3-medium-instruct (128k) on Azure AI Foundry, with the retirement schedule now listing only Phi-4 and no longer publishing an exact date. The model was also delisted from mod…
Reasoning models are deliberately trading world knowledge for reasoning skill, with GLM-5.2 scoring 99.2% on AIME 2026 using about 40 billion active parameters per token, while factual recall remains …
Five researchers posted a preprint describing SEAG, a framework that uses a small local model to replace sensitive entities with aliases before forwarding RAG queries to external large language models…
Microsoft's Visual Studio 2026 18.9, released August 11, adds Copilot thinking effort controls and Ollama local model support, addressing credit spend from usage-based billing. Thinking effort lets us…
DeepSeek-V3.2 defeated Phi-4 in a head-to-head benchmark with an aggregate score of 98.0 to 81.8, an 8–2 task lead with 2 ties, and a 97% confidence verdict, according to an evaluation by gpt-5.4 acro…
Tools4AI, a 100% Java agentic AI framework, can be combined with Ollama's local LLM server to build fully offline AI agents that never send data to third-party APIs. In a tutorial, a developer demonst…
Seldon's Trace Audit v0 clusters LLM traces into compilable programs using hard contract blocks and block-local DBSCAN on a short deterministic feature string, achieving 1.000/0.9999/1.000 pair precis…
Australia should build its own small language models (SLMs) to gain a sovereign competitive advantage, argues a senior technology executive, citing a case where a tourism and cruise business cut opera…
A new study from dessertlab shows that fine-tuned mid-sized open LLMs can outperform GPT baselines in schema-constrained document-level event argument extraction on the MAVEN-ARG benchmark. The approa…
Researchers have proposed a new AI-based health misinformation detector designed for culturally and linguistically diverse (CALD) communities, using Bangla-translated health misinformation data to tes…
A new agent evaluation method that grades models against real server states rather than their own reports found that model size does not determine success, with a 27B reasoning-tuned Qwen derivative a…
A new benchmark measuring the political neutrality of 18 AI models from 12 labs across four regions found that 97 out of 108 measured positions landed left of center, with only xAI's Grok models appro…
Grok-4.3 defeated Phi-4 37.0 to 26.0 in a head-to-head comparison of text generation models, winning on correctness, instruction-following, and precision across four tasks including code generation, e…