Tune your prompt to the model.
Prompting isn't one-size-fits-all. Each model — and each new version — rewards a different style. This is the quick reference for getting the output you actually expect from Claude, ChatGPT, Gemini, and Kimi — in fewer tokens.
Get what you expect
Know each model’s quirks up front — fewer "that’s not what I meant" retries.
Spend fewer tokens
Leaner prompts and the right tier for the job cut cost on repetitive work.
Copy, paste, ship
Real product examples — PRDs, research, roadmaps — not toy prompts.
The models
Today's flagships. Each card carries a “last verified” date — models change weekly.
Claude Fable 5
The detail obsessive — Anthropic's most capable widely released model.
Claude Opus 5
Anthropic's own "start here" default for complex, high-stakes work.
Claude Sonnet 5
The everyday workhorse — and now the default across Claude products.
Claude Haiku 4.5
Fast and cheap — for high-volume, well-defined tasks.
Claude Opus 4.8
Maximum reasoning for complex, long-horizon product work.
Claude Sonnet 4.6
The previous everyday workhorse, replaced as default by Sonnet 5.
GPT-5.6 Sol
OpenAI's frontier workhorse for reasoning, coding, and agentic runs.
GPT-5.6 Terra
The everyday tier — 40% of Sol's price, same window and cutoff.
GPT-5.6 Luna
The cheap, fast tier for volume work — 1/25th of Sol's input price.
GPT-5.5 Thinking
OpenAI's previous deliberate reasoner for hard, multi-step problems.
GPT-5.5 Instant
The previous default ChatGPT model — fast, dependable, broadly capable.
GPT-5.5 Pro
The previous high-effort tier for the most demanding tasks.
Gemini 3.5 Pro
Google's next flagship — announced in June, still not broadly available.
Gemini 3.7 Flash
Google's current workhorse — and the cheapest way into a 1M window from a major provider.
Gemini 3.1 Pro
Stable, capable reasoning — a dependable production choice.
Gemini 3.5 Flash
The previous Flash workhorse, replaced by 3.6 Flash.
Gemini 3.6 Flash
The structure specialist — not the smartest model, but the clearest thinking on the page.
Kimi K3
Ships with product management skills built in — and open weights you can host yourself.
Patterns for real product work
Each pattern comes with model-tuned variants and a note on why a given version is leaner or more reliable — so the token saving is visible, not implied.
Draft & critique a PRD
Turn a rough PRD into something that survives exec scrutiny — and find the weak assumptions before someone else does.
Synthesise user interviews into insights
Turn raw, messy interview notes into evidenced pains and themes — with quotes, not vibes.
Pressure-test a roadmap
Stress-test a prioritisation before you commit — find the assumption that, if wrong, flips the whole order.
Quick competitive scan
Go from a stack of competitor pages to a clean comparison table in one pass — cheaply.
Turn discovery findings into options
Map validated findings into an opportunity-solution tree — and flag which solutions are still untested assumptions.
Write crisp product comms
Exec updates and release notes that lead with the point — fast, in your tone, without the hype.
New to all this? Start with the fundamentals.
Five model-agnostic habits that make every prompt better, before you worry about which model you're on.
Read the best practices