AI Guides › Playbooks
By Nigel Guy · 7 min read
Most people pick a Claude model the way they pick a lift button: whichever is already lit. Others do the opposite and reach for the biggest name every time, then wonder why the bill or the wait is high. Both feel sensible while they are costing you money or quality, because nobody shows you what the choice actually changes.
The rule: start on the cheapest model that could plausibly do the job, raise effort before you raise the model, and move up only when a real output has failed a check you wrote in advance.
A lot of articles still describe the range as "Fable 5.1, Opus 5, Sonnet 5 and Haiku 4.5". Anthropic's own models page, checked on 2026-10-04, lists the current lineup as Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 4.5. Opus 5 and Sonnet 5 are now listed as legacy models, still available. This guide uses the current names and says so where the older ones matter. Model names move quickly, so check the models page before you commit to a workflow.
Specs below are from Anthropic's models overview and pricing pages. Prices are API prices per million tokens (MTok); the £ figures are rough conversions at about 75p to the dollar, so check the current rate.
| Fable 5.1 | Opus 5.5 | Sonnet 5.5 | Haiku 4.5 | |
|---|---|---|---|---|
| Built for | Demanding reasoning, long-horizon agentic work | Long-running agentic coding and knowledge work | Best mix of speed and intelligence | Fastest, near-frontier intelligence |
| Input / output per MTok | $10 / $50 (about £7.50 / £37.50 at time of writing) | $4 / $20 (about £3 / £15) | $2 / $10 (about £1.50 / £7.50) | $1 / $5 (about £0.75 / £3.75) |
| Context window | 1M tokens | 1M tokens | 1M tokens | 200K tokens |
| Max output | 128K | 128K | 128K | 64K |
| Thinking | Adaptive, always on | Adaptive, always on | Adaptive | Extended |
| Reliable knowledge cutoff | Jun 2026 | Jun 2026 | Jun 2026 | Feb 2025 |
| API ID | claude-fable-5-1 |
claude-opus-5-5 |
claude-sonnet-5-5 |
claude-haiku-4-5-20251001 |
Anthropic's own advice is to start most workloads on Opus 5.5, and to move to Fable 5.1 when your tests on Opus 5.5 at higher effort still fall short. That is a sound default for an API build with a budget. For everyday chat work, the card below is cheaper in time and money.
Hypothetical job: you feed in about 50,000 tokens (roughly 28,000 words) and get about 5,000 tokens back. This ignores thinking tokens, which are billed as output and will add to it.
| Model | Input | Output | Total | Approx. £ at time of writing |
|---|---|---|---|---|
| Fable 5.1 | $0.50 | $0.25 | $0.75 | about £0.56 |
| Opus 5.5 | $0.20 | $0.10 | $0.30 | about £0.23 |
| Sonnet 5.5 | $0.10 | $0.05 | $0.15 | about £0.11 |
| Haiku 4.5 | $0.05 | $0.025 | $0.075 | about £0.06 |
Two levers cut this further: the Batch API is 50% off for work that can wait, and prompt caching makes repeat reads of the same material much cheaper. Haiku 4.5 uses an older tokenizer than the others, which Anthropic says produces roughly 30% fewer tokens for the same text, so its gap is slightly different in practice. In the Claude apps you pay a subscription rather than per token, so the cost shows up as usage limits and speed; check the current plan page for the limits on your plan, because I could not verify per-model allowances.
| Where | How |
|---|---|
| Claude apps (web, desktop, mobile) | Click the model name next to the send button, choose a model, or "More models". Changes apply from Claude's next reply, so you can switch mid-chat. Effort and thinking settings sit alongside it. |
| Claude Code | /model opens the picker; /model sonnet switches directly. claude --model opus at startup. Aliases: opus, sonnet, haiku, fable, best, opusplan. /effort high changes effort. |
| Claude API | Set the model field to the ID in the table, for example claude-sonnet-5-5. |
| Cloud platforms | Bedrock, Google Cloud, Microsoft Foundry and Claude Platform on AWS each use their own ID forms; copy them from the models overview, and note those platforms set their own lifecycle dates and pricing. |
In Claude Code, the picker saves your choice as the default when you press Enter and applies it to this session only if you press s. Which models appear in the apps depends on your plan and may differ from this table; check your own picker.
Partly. In Claude Code the opusplan alias uses Opus for planning and then switches to Sonnet to execute, and best picks Fable where available, otherwise Opus. I did not find an official automatic per-message model picker in the chat apps, so assume you choose.
Use this in any chat, on the cheapest model. Fill in the task description, your quality check and your platform.
You are a practical advisor helping me choose between Anthropic's current Claude models: Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 4.5.
My task: [TASK_DESCRIPTION]
How I will judge a good result: [QUALITY_CHECK]
Where I work (Claude app, Claude Code, API or cloud platform): [PLATFORM]
My constraints on speed or budget: [SPEED_OR_BUDGET]
Steps:
1. If any input above is missing or vague, ask me for it before answering.
2. Recommend the cheapest model that could plausibly pass my check, and say what effort level to start at.
3. Give one line on why each other model is not the starting choice.
4. Name the specific sign that would tell me to move one model up, and the sign that I could move one down.
5. Give the exact switching steps for my platform.
Constraints: do not quote prices or limits from memory. Tell me to check Anthropic's models and pricing pages for those. If you are unsure what a model name refers to, say so.
Before you answer, check that your recommendation follows from my stated check rather than from model reputation. Keep it under 200 words.