Search the Practicum

Type two letters · ↑↓ select · Enter open · Esc close

Type at least two letters.

Tools map: what to open and what to pay for

A living Practicum page. Checked: 2026-08-18. Tools, prices, and access change every month. Check the date before you use or pay for anything. This is not advertising or financial advice. Read the terms and use services at your own risk. This page gives a direction and one line on why a tool may help. Do the deeper research yourself.

A note from Chapter 1: output is a finished artifact, such as text, code, a report, or an image. Outcome is the real-world result that changes something. A system has another useful image: the model is the engine, and the harness around it is the car. The harness includes memory, tools, and checks. The tools below are parts of that car, not the whole thing. The book glossary has fuller definitions.

Assistants you can sign into

Tool Why use it Link
ChatGPT (OpenAI) General writing, rewriting, and working a problem through step by step. chatgpt.com
Claude (Anthropic) Long documents and text where tone and voice matter. claude.ai
Gemini (Google) Everyday assistant work, with links into Google's own apps. gemini.google.com
Microsoft Copilot The default if your work already runs on Microsoft 365, because it reads your own mail and files. copilot.microsoft.com
Mistral (France) The strongest European option, with EU data residency. Renamed from Le Chat to Vibe on May 28, 2026. chat.mistral.ai
Grok (xAI) A very large context window and live access to X. Weigh the source quality before you rely on it. grok.com

Each has a free entry level. For the exercises in the books, any one of them is enough.

One pricing change worth knowing. The standalone Copilot Pro subscription at $20 a month can no longer be bought and went out of support on Aug 1, 2026. Microsoft 365 Premium replaced it. If you were paying for Copilot Pro, check what you are on now. Checked Aug 18, 2026.

If you are in the EU and data residency is the deciding factor, Mistral is the one built around that answer rather than retrofitted to it. That is a procurement argument, not a claim that its model is better.

Free models worth knowing

Tool Why use it Link
DeepSeek A capable free model for reasoning. Useful when you are running many drafts. deepseek.com
Qwen (Alibaba) An open model that works in a browser and can also run locally with Ollama. chat.qwen.ai
Perplexity Search with sources. Check the linked sources yourself. perplexity.ai
NotebookLM (Google) Turns your own documents into a structure and an audio overview. Useful for study, procedures, and repackaging content. More detail. notebooklm.google

Assistants store what you send them, and the terms differ by service and by plan. Keep client data, medical records and contracts out of them, and check your employer's policy before your first serious task.

Coding agents: the fastest-moving category

If you write code, or manage people who do, this category changed more in a year than any other on this page. It is worth naming even though most readers will not open these.

Tool Why use it
Claude Code (Anthropic) A terminal agent that edits files across a whole project. By Feb 2026 it was authoring roughly 4% of all public GitHub commits, about 135,000 a day
Cursor An editor built around AI, best for individual focused work with the code in front of you
OpenAI Codex OpenAI's agent, available as a CLI and inside other tools
GitHub Copilot Autocomplete inside the editor. The broad baseline in most companies

The pattern in 2026 companies is not "pick one." It is Copilot deployed to everyone as a baseline, plus an agent like Claude Code adopted by senior engineers for the harder work. Reports of teams running two or three of these together as layers are now common.

The market signal matters more than the tool names. Job postings requiring experience with AI coding tools grew about 340% between Jan 2025 and Jan 2026, while postings for pure implementation roles fell about 17%. That is the output-to-outcome shift from Chapter 2, visible in hiring data: the value moved to the person who directs and reviews the work, not the person who types it.

A working move: prototype cheaply, finish carefully

You do not need to do every task with the most expensive model. Use a cheaper capable model for drafts, research, and large repeated runs. Reserve the expensive one for the final polish and check.

One pricing example, checked against the official price list on Aug 18, 2026. The DeepSeek V4 Pro API lists $1.98 per million output tokens off-peak and $3.96 at peak. Uncached input is $0.66 off-peak, and a cache hit is $0.022. Cached input is therefore about 30 times cheaper than uncached input, and there is a second, cheaper model, V4 Flash, at roughly a third of the Pro output price.

Two things are worth taking from that, and neither is the number itself. First, the prices moved: this page previously carried figures that were less than half the current ones, which is what a monthly check is for. Second, the vendor now charges by time of day, with off-peak rates at half the peak rate. If you run large batch jobs, when you run them is now a cost decision.

This is a way of working, not a model recommendation. Prices and terms change.

Start here

Open one assistant from the first table and take a real task to a finished result. Do not chase "the best" tool. What you can access today, used properly, beats a better tool you never open.

Reading from Russia? The leading assistants are not directly available there, and the Russian version of this page maps the local routes: tools map, in Russian.

Update rule: every month. Mark out-of-date rows with a date rather than deleting them. The map by system layer, such as Thinking, Knowledge, and Tools, is here. Curated channels and courses: recommended resources. Version: 2026-08-18.

Sources checked on 2026-08-18

Features, prices, and access change. Check the primary sources:

Prices go stale quickly. Keep the "Checked" date and the caveat visible.

Tools map: what to open and what to pay for