Categories

Large Language Models News

A comment trigger, reasoning dial, and pull request checkpoint connect through a cloud agent workspace

GitHub gives Copilot cloud agent comment triggers and reasoning controls

A compact local model core links a laptop, phone, and edge device inside a privacy boundary

Liquid AI releases LFM2.5-2.6B for on-device agents

An autonomous vehicle sensor module links a cloud reasoning model to in-vehicle compute inside a simulated road lab

NVIDIA makes Alpamayo 2 Super available for commercial robotaxi development

Metrics, logs, and traces converge toward an unresolved root-cause node on an abstract observability dashboard

ORCA-bench shows coding agents are not ready for on-call root cause analysis

A new benchmark tests AI agents on on-call root cause analysis with metrics, logs, traces, code, and 1,079 incident tasks.

The AI Feed Desk

By The AI Feed Desk

Legacy model blocks move across a routing bridge toward a long-context AI system beside an abstract deadline marker

Kimi API sunset turns K3 migration into an August deadline

Kimi's model documentation now warns that K2.5 and moonshot-v1 are closed to new users and that the older platform sunsets on August 31.

The AI Feed Desk

By The AI Feed Desk

A luminous Q-shaped data path runs between cloud infrastructure and modular open model weights

Alibaba releases Qwen3.8-Max with QwenCloud access and open weights next week

Qwen3.8-Max puts Alibaba back into the frontier-model fight with a large coding-focused model available through QwenCloud now and open weights promised next week.

The AI Feed Desk

By The AI Feed Desk

Layered pull request cards rise from one branch into separate review panels before merging

GitHub stacked pull requests give coding agents smaller review units

GitHub put stacked pull requests into public preview, letting developers and coding agents split large changes into ordered, reviewable layers.

The AI Feed Desk

By The AI Feed Desk

Mathematical proof pages become verified geometric certificates on a quiet research desk

OpenAI publishes ten AI-generated math and theoretical CS advances

OpenAI published ten AI-generated results in mathematics and theoretical computer science with Lean 4 formalizations in a public repository.

The AI Feed Desk

By The AI Feed Desk

Three model-routing lanes on an operations table show low-cost, balanced, and fast AI work paths

OpenAI cuts GPT-5.6 Luna and Terra prices

OpenAI cut GPT-5.6 Luna pricing by 80%, Terra pricing by 20%, and added a faster Sol API mode for latency-sensitive work.

The AI Feed Desk

By The AI Feed Desk

A compact inference chip connects terminal panels and tool-call paths for a coding agent API

DeepSeek ships V4-Flash public beta with Responses API support

DeepSeek moved V4-Flash into public beta with native Responses API support, Codex configuration docs, and a new leaderboard entry for V4-Flash-0731.

The AI Feed Desk

By The AI Feed Desk

A code review lens checks pull request cards using read-only context connectors and team skill folders

GitHub makes Copilot review skills and MCP generally available

GitHub moved Copilot code review agent skills and MCP to general availability while preparing default model enablement for enterprise customers.

The AI Feed Desk

By The AI Feed Desk

A coding-agent model picker routes a repository task to Grok 4.5 with policy and usage controls visible beside it

GitHub adds Grok 4.5 to Copilot's model picker

GitHub is rolling out xAI's Grok 4.5 across Copilot surfaces, with enterprise access controlled by an admin policy that is off by default.

The AI Feed Desk

By The AI Feed Desk

A managed Gemini agent runs tool calls through policy hooks, budget gauges, and scheduled worker lanes inside a cloud sandbox

Google turns Gemini Managed Agents into scheduled, governed workers

Google updated Gemini API Managed Agents with Gemini 3.6 Flash as the default, environment hooks, token budgets, scheduled triggers, and free-tier access.

The AI Feed Desk

By The AI Feed Desk

A Claude Opus 5 model control room routes coding, business, and research agent tasks through cost and safety controls

Claude Opus 5 becomes Anthropic's everyday agent model

Anthropic says Claude Opus 5 is available on Claude and the API at Opus 4.8 pricing, with stronger agentic coding and professional-work performance.

The AI Feed Desk

By The AI Feed Desk

Developer work lanes compare coding, agent tasks, reasoning effort, and billing controls for Gemini 3.6 Flash in Copilot

Gemini 3.6 Flash rolls into GitHub Copilot

GitHub says Google's Gemini 3.6 Flash is rolling out in Copilot with configurable reasoning effort and parallel tool use.

The AI Feed Desk

By The AI Feed Desk

A security operations console rotates token keys into a vault beside a dataset processing pipeline

Hugging Face says an autonomous agent breached production infrastructure

Hugging Face disclosed a July 2026 production incident it says was driven by an autonomous AI agent system and recommends token rotation.

The AI Feed Desk

By The AI Feed Desk

An open model engine sits on a lab table with layered context bands and a release-calendar marker

Moonshot launches Kimi K3 as an open 2.8T model

Moonshot's Kimi K3 pairs a 2.8T-parameter model, 1M-token context, API access, and a promised July 27 weight release.

The AI Feed Desk

By The AI Feed Desk

Hot MoE experts move between SRAM, HBM, and shared DRAM tiers on a chiplet package

HCRMap targets hot-expert bottlenecks in MoE inference

A July 13 paper proposes HCRMap, a pressure-aware residency system for hot MoE experts across 3.5D chiplet memory tiers.

The AI Feed Desk

By The AI Feed Desk

A central managed agent node connects to four external tool and background task endpoints

Google adds background tasks and remote MCP to Gemini Managed Agents

Google expanded Managed Agents in the Gemini API with asynchronous background execution, remote MCP servers, custom functions, and credential refresh.

The AI Feed Desk

By The AI Feed Desk

A productivity workspace connects three app tiles to a central approved model marker

GPT-5.6 becomes Microsoft 365 Copilot's preferred model

OpenAI says GPT-5.6 is now the preferred model in Microsoft 365 Copilot across Word, Excel, PowerPoint, Chat, and Cowork.

The AI Feed Desk

By The AI Feed Desk