Z.ai released the 753B-parameter GLM-5.3 flagship's weights on Hugging Face on August 28, 2026, under a bespoke licence named 'glm-5.3', while the 320B GLM-5.3-Flash model shipped two days earlier und…
Fal, an inference-infrastructure company, announced on August 27, 2026, that its new H3 Max model generates five seconds of AI video with audio in under three seconds, claiming roughly 35x the through…
A new reference taxonomy from an unnamed source normalizes AI video spec numbers, revealing that '40 seconds' and '4K' claims collapse into five distinct mechanisms across duration and resolution. Amo…
AI agents can silently return wrong answers when they read from stale data copies, and the staleness budget should be set per question rather than per system, according to a guide that outlines a fram…
PostgreSQL row-level security is the recommended enforcement point for AI agents querying business data, as UI-level scope checks fail when agents compose their own queries, according to a guide from …
A general-purpose coding agent such as Claude Code, Codex CLI, or Gemini CLI can outperform single-purpose marketing AI tools for data and file tasks, according to a playbook that identifies five tran…
Google DeepMind launched Gemini Omni 1.1 Flash on August 27, 2026, enabling video scenes up to 40 seconds via 10-second extension passes and a 360p draft tier priced at roughly one-third of the 720p r…
Z.ai revealed on August 26, 2026, that the anonymous model `stealth/ox-alpha` on OpenRouter was GLM-5.3-Flash, publishing MIT-licensed weights the same day. The named listing, `z-ai/glm-5.3-flash`, en…
A 2026 census of eleven hosted model APIs found that free AI tiers often carry different data terms than paid tiers, with Google and Mistral explicitly stating their free tiers use prompts for trainin…
On August 26, 2026, OpenRouter's anonymous 'stealth/ox-alpha' listing was revealed to be Z.ai's GLM-5.3-Flash, with MIT-licensed weights published the same day, prompting a re-verification job for tea…
A new reference from Artificial Analysis, based on vendor documentation fetched August 26, 2026, finds that image and video input billing across AI APIs is highly inconsistent, with the same 1024×1024…
Evaluating an anonymous AI model requires separating measurable behavior from unknowable identity, as stealth listings like ox-alpha on OpenRouter offer no named entity for retention, liability, or de…
As of August 25, 2026, OpenAI, Azure OpenAI, Google Vertex AI, and Mistral all publish a 10% surcharge (1.1× multiplier) for data residency on AI inference, while AWS Bedrock shows no residency line i…
Apple's M5 Ultra Mac Studio, announced August 25, 2026, offers up to 512GB of unified memory at 1.2TB/s, enough to hold trillion-parameter-class open-weight models like DeepSeek-V4-Pro (1.6T params, ~…
Anthropic and OpenAI price cache reads at 0.1x the base input rate, and Anthropic states caching pays off after one read (5-minute tier) or two (1-hour tier), but a multi-hour agent run can keep spend…
A census of eight API vendors and four coding-CLI surfaces found that the dominant documented behavior when hitting an AI spend cap is a hard stop via HTTP 429, 402, or 400, not a graceful degrade, ac…
As of August 25, 2026, AI video generation vendors quote prices in four incompatible units, making direct cost comparison impossible, according to a reference that normalizes published rates into a si…
A census of nine AI coding CLIs, dated August 24, 2026, found that first-run behaviors vary widely and are often undocumented, with a confirmed Cursor CLI flaw allowing arbitrary command execution bef…
LLM tokenizer variance means identical dollar-per-million token prices can produce materially different bills, because each vendor's tokenizer segments the same input into different token counts. Anth…
A new public ledger documents 27 dated Model Context Protocol (MCP) security incidents and protocol revisions from April 2025 through August 24, 2026, with 25 incidents and 2 protocol revisions. Comma…