The Field Guide Living document · updated 08 Jul 2026

Which AI,
for what, and why.

A plain-language, always-updated cheat sheet to the major AI assistants — the models under the hood, what each is genuinely good at, which "thinking level" to use, and when to reach for something else. No hype, no shilling.

5
Tools mapped
8
Fields per tool
7
Role playbooks
Weekly
Update cadence
100%
Source-cited

New here? Search a task like "write an email", "fix HTML", or "make a spreadsheet" — or tap a role below to see exactly which tool, which model, and a ready-to-paste prompt. Every card opens to a full spec.

Filter by task

All tools Writing Coding Research Office / Docs Images / Video Everyday

Or filter by who you are

🎓 Student 🍎 Teacher 🎨 Content creator 💻 Developer ✉️ Support / Email 📊 Manager 🔍 Researcher

01 Overview

Anthropic's assistant, widely praised for the quality of its writing and its steady reasoning over long documents and codebases. It leans helpful, honest and cautious — a favourite for drafting, editing, coding and analysis.

02 Models powering it

Claude Fable 5Most capable · 1M-token context
Claude Opus 4.8Flagship for complex work
Claude Sonnet 5Default on Free & Pro · fast + agentic
Claude Haiku 4.5Cheapest & fastest · routine tasks

03 The effort dial (layman's guide)

When you pick a model, Claude lets you set how hard it thinks — plus a Thinking toggle that shows its reasoning.

  • Low / Medium — quick, routine tasks; stretches your usage further
  • High — the best all-round balance (the recommended default)
  • Max — the hardest, get-it-right work; expect slower replies

Source: Claude Help Center

04 Best use cases

  • Long-form writing: reports, articles, briefs, structured docs
  • Coding & multi-file refactors (chat or Claude Code)
  • Reading and analysing long documents
  • Agentic, multi-step tasks that run unattended

05 Where to access

Web, desktop & mobile at claude.ai · the API · Claude Code (for developers) · plus Claude inside Chrome, Excel and PowerPoint.

06 Pricing tiers

Free · Sonnet 5Pro · ~$20/moMax · ~$100–200/moAPI · Sonnet 5 $3/$15*

*Intro API $2/$10 per 1M tokens to 31 Aug 2026; Haiku 4.5 ≈ $1/$5. Source: Claude docs

07 Prompting tips + example

  • Be explicit and detailed; give a good and bad example
  • Use XML-style tags to separate parts of a prompt
  • Ask it to "think step by step" for harder problems
Try this"You are my editor. Tighten the draft below to 200 words, keep my voice, and flag anything unclear. <draft>…</draft>"

08 Limits & when NOT to use it

  • Fixed knowledge cutoff — it searches the web for recent facts
  • Rate limits bite on Free / lower tiers under heavy use
  • Not for live in-editor autocomplete, Office-embedded work, or image generation
The top Fable / Mythos models had a brief export-control pause in June 2026, restored 1 Jul 2026; Opus, Sonnet and Haiku were unaffected.

01 Overview

OpenAI's assistant and the household name in AI. The jack-of-all-trades: quick answers, writing, coding, image generation, voice and deep research in one place, with the largest ecosystem of plugins and integrations.

02 Models powering it

GPT-5.3 InstantDefault for all incl. Free · fast
GPT-5.5 ThinkingPaid reasoning for harder tasks
GPT-5.5 / 5.5 ProFlagship · research-grade
Codex / Codex-SparkAgentic coding & IDE autocomplete

A newer GPT-5.6 family (Sol/Terra/Luna) is in limited API & Codex preview — not yet in ChatGPT. Source: OpenAI

03 The effort dial (layman's guide)

ChatGPT keeps it simple with two modes (it can auto-switch):

  • Instant — fast answers for everyday questions
  • Thinking — slower, more careful reasoning for hard problems
  • Pro mode — highest effort on Pro tiers; note it disables Memory, Canvas and image gen while running

04 Best use cases

  • Everyday questions, drafts, brainstorming, learning
  • Image generation and editing
  • Voice conversations and hands-free use
  • Agentic coding via Codex; deep research reports

05 Where to access

Web at chatgpt.com · iOS & Android · desktop apps · the API · and Codex for developers.

06 Pricing tiers

Free · GPT-5.3, ads in USGo · ~$8/moPlus · $20/moPro · $100 & $200/mo

Plus caps Deep Research at ~10/mo; the 1M-token context sits on Pro $200. Source: ChatGPT pricing

07 Prompting tips + example

  • Set Custom Instructions so you don't repeat context
  • Pick the mode: Instant for speed, Thinking for hard tasks
  • For images: state subject, style, mood, composition
Try this"Create a clean, minimal logo for a coffee cart called 'Slow Pour' — warm palette, flat vector style, on a white background."

08 Limits & when NOT to use it

  • Chats may train models unless you opt out — mind confidential work
  • Deep Research / Thinking have quota caps on lower tiers
  • Less suited to Office-embedded work grounded in your company files

01 Overview

Microsoft's assistant, embedded inside the apps most offices already live in. Its superpower isn't raw model quality — it's context: grounded in your emails, files and meetings via Microsoft Graph. Note "Copilot" is a brand covering several products.

02 Models powering it

OpenAI GPT-5.x5.3 Instant + 5.4 Thinking in M365
Microsoft modelsPhi & MAI for certain tasks
Model choiceAnthropic + OpenAI in Excel Agent Mode
GitHub CopilotOpenAI, Anthropic & Google models

Copilot routes across several models rather than running one. Source: Microsoft 365 Copilot

03 The effort dial (layman's guide)

You don't pick a "thinking level" here — Copilot chooses the model for you. What you control is context: point it at the right file, email or cell range, and use Agent Mode in Word/Excel for multi-step jobs.

04 Best use cases

  • Drafting & summarising in Word, Outlook, Teams
  • Excel: formulas, analysis and insights on your sheets
  • PowerPoint decks from a document or brief
  • Coding in your IDE via GitHub Copilot

05 Where to access

Free Copilot chat on web, Windows & mobile · inside Word / Excel / PowerPoint / Outlook / Teams with a paid licence · Edge · GitHub Copilot in code editors.

06 Pricing tiers

Free · Copilot ChatM365 Premium · ~$19.99/moBusiness · ~$21–25/seatEnterprise · ~$30/user/mo

Business/Enterprise sit on top of a paid M365 base licence. M365 Premium replaced the old Copilot Pro for individuals. Source: Microsoft

07 Prompting tips + example

  • Point at specific files, emails or ranges
  • Work inside the app where the data lives
  • Name the format: "a 5-slide deck", "a 3-bullet summary"
Try this (in Excel)"Analyse this sales table, add a column for month-over-month growth %, and summarise the three biggest movers."

08 Limits & when NOT to use it

  • Value depends on you living in the Microsoft ecosystem
  • Paid tiers require an underlying M365 licence — cost is stacked
  • Only sees data you already have permission to view
  • Not ideal if your stack is Google / Slack / Jira, or you just want a standalone chatbot

01 Overview

Google's assistant, and the natural pick if you live in Gmail, Docs and Google Search. It has one of the largest context windows around, a genuinely generous free tier, and standout multimodal features including video generation.

02 Models powering it

Gemini 3.1 ProDeep reasoning · 1M-token context
Gemini 3.5 FlashFast default · strong at coding
Gemini Deep ThinkHardest problems (Ultra tier)
Nano Banana / VeoImage & video generation

03 The effort dial (layman's guide)

  • Flash — the fast default; great for everyday tasks
  • Pro — deeper reasoning for harder work and long documents
  • Deep Think — extended reasoning for the toughest problems (Ultra only)

04 Best use cases

  • Deep Research reports across many live sources
  • Working inside Gmail, Docs and Sheets
  • Very long documents (up to ~1,500 pages of uploads)
  • Image and video generation

05 Where to access

Web at gemini.google.com · iOS & Android · inside Google Workspace (Gmail, Docs, Sheets) · Android system assistant · API via Google AI Studio.

06 Pricing tiers

Free · Flash + daily ProAI Plus · ~$7.99/moAI Pro · ~$19.99/moAI Ultra · from ~$99.99/mo

Free tier includes Flash, daily Pro access, image gen and some Deep Research. Source: Google

07 Prompting tips + example

  • Lean on it inside Docs/Gmail where your content already is
  • Upload long files and ask for structured summaries
  • Use Deep Research for anything needing many sources
Try this"Deep Research: compare the top 3 project-management tools for a 10-person team — pricing, key features, and best fit. Cite sources."

08 Limits & when NOT to use it

  • Best value is realised inside the Google ecosystem
  • Some top features are region-limited or Ultra-gated
  • Occasional feature-parity bugs reported by paid users in early 2026
  • If you want Microsoft-Office-native work, use Copilot instead

01 Overview

A search-and-answer engine that reads the live web and hands you a concise, cited answer. Rather than one house model, it lets you route queries to several frontier models — its edge is trustworthy, sourced research.

02 Models powering it

SonarNative model (Llama-based) · fast search
Frontier modelsGPT-5.x, Claude Opus, Gemini — switchable (Pro)
Model CouncilRuns 3 models at once (Max)
Perplexity ComputerOrchestrates many models for agentic work (Max)

03 The effort dial (layman's guide)

  • Quick Search — a fast, sourced answer to a question
  • Pro Search — multi-step reasoning for deeper questions
  • Deep Research — visits 100+ pages and writes a structured, cited report

04 Best use cases

  • Fact-checking and "what's the latest on…" questions
  • Market scans, due diligence, competitor research
  • Anything where you need sources you can click
  • Stress-testing a conclusion across models (Model Council)

05 Where to access

Web at perplexity.ai · iOS & Android · the Comet browser · a Chrome extension · and the Sonar API for developers.

06 Pricing tiers

Free · ~5 Pro searches/dayPro · $20/mo ($10 students)Max · $200/moEnterprise · from ~$40/seat

Pro unlocks unlimited Pro Search and model switching. Source: Perplexity

07 Prompting tips + example

  • Ask specific, factual questions — it rewards precision
  • Request a date range or "cite sources" explicitly
  • Use Deep Research for anything report-shaped
Try this"What changed in EU AI regulation in the last 3 months, and what does it mean for a small SaaS company? Cite recent sources."

08 Limits & when NOT to use it

  • Not built for long-form creative writing or heavy coding
  • It's an answer engine, not a workflow tool — it can't act inside your apps
  • Free tier caps Pro searches quickly
  • For drafting or coding, reach for Claude or ChatGPT instead
No tools match that search or filter. Try clearing it.

How hard should the model think?

The single most useful skill — and nobody explains it. Match the setting to the job.

Quick & simple

Fast · cheap · everyday

Short, obvious tasks where "good and fast" beats "perfect and slow." Rewrite a sentence, format a list, quick fact.

Claude Low
ChatGPT Instant
Gemini Flash

Everyday work

Balanced · the default

Most real tasks: summarise a report, draft an email, plan a week, write a first draft. Leave it here 90% of the time.

Claude High
ChatGPT Instant → Thinking
Gemini Pro

Hard & high-stakes

Deep · slower · get it right

Multi-step logic, tricky code, contracts, big decisions where a subtle error matters. Turn thinking on.

Claude Max + Thinking
ChatGPT Thinking / Pro
Gemini Deep Think
📌 Rule of thumb: start on the default. Turn the dial up only when an answer feels shallow or the task is genuinely complex; turn it down for speed on simple things. Higher isn't better — it's just slower and pricier when the task didn't need it.

Which tool for what

The 30-second decision.
The job
Reach for
Why
Long report or polished writing
Claude
Best structure and voice over long output; big context for source material.
Quick everyday question
ChatGPT / Gemini
Fast defaults, broad general knowledge; Gemini's free tier is generous.
Working inside Word / Excel / Outlook
Copilot
Lives in the app and reads your own files, emails and meetings.
Working inside Gmail / Google Docs
Gemini
Native to Google Workspace, with your content right there.
Sourced research you can cite
Perplexity
Answer engine with clickable citations on every claim.
Generating or editing images
ChatGPT / Gemini
Both have strong built-in image models; Gemini also does video.
Big multi-file coding task
Claude / ChatGPT
Claude Code and OpenAI Codex both run long, autonomous coding jobs.
Autocomplete as you type in an IDE
Copilot
GitHub Copilot leads on real-time, in-editor completion.
Answering from your company's data
Copilot
Grounded in Microsoft Graph across your org's files and messages.

The role playbook

Pick your job → get the tool, the model setting, and a prompt to paste.

The prompt kit

A formula that works everywhere — plus a prompt that writes prompts for you.

The universal prompt formula

Stack these five pieces and almost any model gives a better answer.

Role+Task+Context+Format+Constraints
Worked example"You are a careful copy editor (role). Rewrite the email below (task) for a frustrated customer whose order is late (context). Give me 2 versions, under 120 words each (format). Warm but not groveling; no refund offer (constraints)."

The prompt that writes prompts

Can't word it yourself? Paste this into any chat model and let it interview you.

You are a prompt engineer. I want to accomplish this goal: [describe your goal in one line] Before writing anything, ask me 3–5 short questions to fill in what you're missing (audience, tone, format, length, constraints). Wait for my answers. Then write one clear, optimized prompt I can paste into an AI assistant to get the best result — and briefly note which "thinking level" to use.

Last updated log

Every change is dated — that's how you know it's alive.
08 Jul 2026
v2 — big update. Switched to light theme. Added Gemini and Perplexity (now 5 tools). New: keyword-powered search, task + persona filters, the "how hard should the model think?" explainer, a 7-role playbook, a prompt kit with a meta-prompt, and a feedback panel.
08 Jul 2026
v1 — initial build. First three entries (Claude, ChatGPT, Copilot) plus the decision matrix.
— next —
Planned: a domain repository (prompt templates & notes per sector), a jargon glossary, and Grok + open-source models (Llama, Mistral).
Copied to clipboard