7 minutes to master any new AI
Fresh, hands-on tutorials for every new AI release.
New model dropped? New tool launched? We test it, screenshot it, and turn it into a 7-minute guide you can actually use — never a press release rewrite.
Hand-tested by a builder · Live within 24h of every release · 119 guides & briefs shipped
Tutorials
view all →How to do it — we test each release and write the step-by-step guide you can follow.
-
How to Switch to Claude Opus 5 in Cursor, Claude Code & the API
Claude Opus 5 shipped July 24, 2026 at the same $5/$25 price as Opus 4.8. Here's how to enable it everywhere — the exact claude-opus-5 model ID, the Claude Code and Cursor steps, the Bedrock/Vertex strings, the effort default, and the silent Opus 4.8 fallback on flagged requests that trips up security tooling.
-
How to Use Gemini 3.6 Flash (and When to Pick Flash-Lite Instead)
Google shipped three Flash models on July 21, 2026: Gemini 3.6 Flash (general production — $1.50/$7.50 per 1M, 17% fewer output tokens than 3.5 Flash with real quality gains), Gemini 3.5 Flash-Lite (high-volume, $0.30/$2.50, 350 tokens/sec), and Gemini 3.5 Flash Cyber (security, gated to partners — not GA). This guide covers which model to pick, the API model ids, pricing, the efficiency and benchmark gains, where each is available, and how to switch from 3.5 Flash.
-
How to Access Qwen 3.8 Max Preview Today — What's Confirmed, What Alibaba Hasn't Said, and When Weights Drop
Alibaba previewed Qwen3.8-Max on July 19, 2026 — a claimed 2.4-trillion-parameter multimodal model (text, images, video, documents), positioned 'second only to Fable 5.' But it's a preview: the active-parameter count, benchmarks, license, and open-weights date are all undisclosed, and there's no HuggingFace repo yet — so you can't run it locally. This guide separates what's actually confirmed from the gaps, shows how to access the Max Preview today (Token Plan / Coder, at 10% of standard rate, OpenAI- and Anthropic-compatible), and explains why an open-weight Max flagship would matter.
-
How to Use Grok Build (xAI's Open-Source Terminal Coding Agent)
xAI open-sourced Grok Build (the `grok` CLI) under Apache 2.0 — a Rust terminal coding agent that edits files, runs shell commands, searches the web, and manages long-running tasks as a full-screen TUI, headlessly for CI, or embedded in editors via the Agent Client Protocol. This guide covers installing it (one curl command), authenticating, the three run modes, the features that matter (MCP servers, skills, hooks, sandboxing), building from source, and an honest note on auditing what any AI coding CLI sends over the wire.
-
How to Use Kimi K3 Today (API + Kimi Code) — the 2.8T Open Model, Weights Dropping July 27
Moonshot's Kimi K3 launched July 16, 2026 — a 2.8-trillion-parameter open Mixture-of-Experts model with Kimi Delta Attention, a 1M-token context, and multimodal (text + image) input. The open weights don't drop until July 27, so you can't run it locally yet — but you can use it now via the Moonshot API (model id kimi-k3, OpenAI-compatible), Kimi Code, kimi.com, or OpenRouter. This guide covers exactly how to access K3 today, the pricing ($3/$15 per 1M, $0.30 cache-hit), the benchmarks, and what to expect when the weights land.
-
How to Run Inkling Locally (Thinking Machines' 975B Apache-2.0 Multimodal Model)
Inkling is Thinking Machines' new open-weights model (released July 15, 2026) — a 975B-total / 41B-active Mixture-of-Experts model under Apache 2.0, multimodal (text, vision, audio), with a 1M-token context and strong benchmarks (77.6% SWE-bench Verified, 97.1% AIME 2026). But 975B parameters means the local route is heavy: the smallest usable quant needs ~290 GB of memory. This guide gives the honest hardware reality, then the real ways to run it — llama.cpp with Unsloth's GGUF quants, the official vLLM / SGLang path, the lighter Inkling-Small (276B/12B) variant, and the hosted APIs for everyone else.