Understand LLMs well enough to make better decisions.

Diff · 2026-07-09

OptiLLM: The Drop-In Proxy That Makes Any LLM Reason Better — No Retraining Required

An open-source, OpenAI API-compatible proxy that wraps LLM calls with 20+ inference-time optimization techniques — chain

Diff · 2026-07-09

GPT-5.6 Public Launch Today: Sol, Terra, Luna Family Revealed

OpenAI releases GPT-5.6 in three tiers — Sol, Terra, Luna — with benchmark scores up to 91.9%, pricing from $1/M tokens,

Diff · 2026-07-09

Anthropic Raises $65 Billion in Series H at $965 Billion Valuation — Largest AI Funding Round, Surpasses OpenAI

On May 28, 2026, Anthropic closed a $65 billion Series H financing round, achieving a $965 billion post-money valuation

Diff · 2026-07-09

Claude Sonnet 5 vs GPT-5.5: Benchmarks, Pricing, and the Mid-Tier Sweep

Claude Sonnet 5 beats GPT-5.5 across every directly comparable benchmark while costing 40–50% less. Here's the full brea

Diff · 2026-07-09

Gemini 3.5 Pro Cleared for July Launch While Fable 5 and GPT-5.6 Remain Restricted

Google's Gemini 3.5 Pro ships freely while competitors face regulatory blocks — the first real-world case of AI regulato

Diff · 2026-07-09

Deploying Distributed AI Inference: Blueprints & Troubleshooting

Seven practical deployment blueprints for distributed AI inference on vLLM, llm-d, and Red Hat OpenShift AI — matched to

Diff · 2026-07-09

Llama 4 vs Qwen 3.5 vs Mistral Large 3: The Best Open-Weight LLM in July 2026

A head-to-head comparison of the three strongest open-weight LLM families — Llama 4 Scout, Qwen 3.5/3.6, and Mistral Lar

Diff · 2026-07-08

NVMe KV Cache Offloading for LLM Inference: Serve 10x More Users on One H100

Turn fast local storage into a third tier alongside GPU memory — serving 10x more concurrent users with 2x throughput an

Evals · 2026-07-08

How LLM benchmarks work and what they miss

LLM benchmarks measure capability within controlled datasets, but they fail to account for deployment reality: latency,

Diff · 2026-07-07

Mistral AI Unveils New 'Fat' Open-Weight Model Family — June Tease Becomes July Early Access

Mistral AI's new fat open-weight model family expands on Large 3's MoE design, targeting Anthropic's Sonnet 5 with more

Diff · 2026-07-07

Claude Sonnet 5 Is Now the Default for Free and Pro — What the Model Shift Means

Claude Sonnet 5 is Anthropic's default for Free and Pro users — near-Opus performance, agentic free-tier access, and a n

Diff · 2026-07-07

Claude Sonnet 5 vs Opus 4.8: Agentic Pricing and Cost-Performance Tradeoffs

Claude Sonnet 5 delivers 80–90% of Opus 4.8's agentic coding quality at 20–40% of the cost. Here's how to choose, combin

Diff · 2026-07-07

OpenAI Retires GPT-4.5 and o3: What Developers Need to Know

OpenAI is retiring GPT-4.5 and o3 from ChatGPT and the API, accelerating its model lifecycle from 18 months to 6. Here's

Diff · 2026-07-07

Claude Sonnet 5 Tokenizer Benchmark: Token Count, Cost, and Performance Impact

Sonnet 5's new tokenizer produces 30–42% more tokens at the same per-token price — here's what the benchmarks, cost audi

Diff · 2026-07-07

OpenAI Retires Version Numbers for Sol, Terra, and Luna Tier Naming System

OpenAI abandons traditional model versioning in favor of durable capability tiers — Sol, Terra, and Luna — under GPT-5.6

Diff · 2026-06-29

Google Unveils Open Knowledge Format v0.1 — The Missing Standard for AI Agent Knowledge

Google Cloud published OKF v0.1, an open, vendor-neutral specification for representing knowledge that AI agents can rea

Diff · 2026-06-29

Anthropic Gets Partial Mythos 5 Access After $20M Compliance Overhaul — The Selective Restoration That Redefines Frontier AI Access

Fifteen days after the US Commerce Department ordered Anthropic to suspend Claude Mythos 5 and Fable 5 globally, the gov

Diff · 2026-06-29

OpenAI's Confidential IPO S-1: $25B Revenue, But Losing $1.22 Per Dollar Earned

OpenAI confidentially filed its S-1 with the SEC on May 22, 2026, targeting a September listing above $1 trillion. The f

Diff · 2026-06-29

OpenAI Shifts to Three-Tier Pricing with GPT-5.6 — Named Tiers, Government-Only Limited Preview, and Ultra Subagent Mode

On June 26, 2026, OpenAI previewed GPT-5.6 in a bold new strategy: instead of a single model with a 'reasoning strength

Diff · 2026-06-29

Microsoft Launches Seven MAI Models at Build 2026 — Including First Proprietary Reasoning Model

At Microsoft Build 2026, Microsoft AI unveiled its first in-house MAI model family spanning seven new models across reas

Cache

Concepts, explainers, comparisons and mental models for the things AI people say too quickly.

Run

Practical workflows: choosing a model, testing RAG, pricing a feature, or running an agent safely.

Diff

Short context pieces that answer what changed, who should care, and what evidence is still missing.

How it is made

The site is transparent about AI-assisted writing, model-labelled writer and editor roles, and human review rather than pretending prose appears by magic.