Anthropic · Released, not on our provider
How to prompt Fable 5.2: Anthropic’s advice with example prompts (2026)
How to prompt Fable 5.2: Anthropic’s own guide says to state the task, format and reason clearly, tell it to finish the whole task, say who owns the code, set effort as your task needs, and drop old "think less" lines, and the tables below pair each tip with an example prompt of ours.
Use it now
Chat
This chat starts with Claude Fable 5.2, not Fable 5.2 prompting.
≈ 11 credits per reply · See all models
New accounts get 5 free credits once, valid for 7 days. They work on the lower-priced models here; the priciest ones need credits you bought.
Start chatting — 5 free credits- Maker
- Anthropic
- Type
- Text
- Released
- No date given
- OpenRouter id
- None
Checked
Fable 5.2 prompting: facts, with sources
- Anthropic has a page of its own, "Prompting Claude Fable 5.2", for patterns specific to the model: effort, early stopping, approvals, code ownership, progress updates, subagents, caution that builds up, prompts written for earlier models, final reports and refusals. It sends you to its general Prompting best practices page for techniques that apply to all current Claude models.Source: platform.claude.com · checked 2026-11-19
- Anthropic’s effort page lists five levels for Claude Fable 5.2: low, medium, high, xhigh and max. The API default is high. It advises starting at high, moving down to medium or low where your own tests show quality holds, using xhigh only for the hardest long-running work where you measured a gain, and keeping max for measuring the ceiling.Source: platform.claude.com · checked 2026-11-19
- Anthropic’s model page says adaptive thinking is always on and cannot be turned off, so effort is the control for thinking depth. It says to omit temperature, top_p and top_k, and that forced tool use returns a 400 error. These are API settings: the chat on this site does not expose them.Source: platform.claude.com · checked 2026-11-19
- Anthropic’s launch page says Medium effort is the default in the Claude apps and Claude Code, while its docs give high as the default on the API. It reports its own tests where Fable 5.2 at lower effort matches other models at higher effort; those are Anthropic’s figures, not tested by us.Source: anthropic.com · checked 2026-11-19
- Anthropic says Fable 5.2 follows instructions closely. A line such as "answer directly without deliberating" does not stop it thinking at medium effort and above, a hard cap such as one instruction per sentence in 20 words or fewer can make its prose choppy, and an instruction containing "never" is treated as absolute.Source: platform.claude.com · checked 2026-11-19
- Fable 5.2 runs safety classifiers that can decline a request. Anthropic names six refusal categories: cyber, bio, frontier_llm, reasoning_extraction, general_harms and null. It says requests asking the model to reproduce its internal reasoning in the response text fall under reasoning_extraction, and tells API users to remove such instructions and read the thinking blocks instead.Source: platform.claude.com · checked 2026-11-19
- Anthropic says that for most workloads you should start with its latest Claude Opus model and move to Claude Fable 5.2 when your own tests on Opus at higher effort still fall short. Each current Claude model has its own prompting page, so advice tuned for one model should be re-tested on another.Source: platform.claude.com · checked 2026-11-19
- Anthropic’s model page lists Claude Fable 5.2 at $8 per million input tokens and $40 per million output tokens, cache reads at $0.20 per million, and a 50 percent discount on the Batch API.Source: platform.claude.com · checked 2026-11-19
- Claude Fable 5.2 is on OpenRouter’s list as anthropic/claude-fable-5.2 at $8 / $40 per million tokens (checked November 19, 2026), so you can try these prompts in this site’s chat. The chat does not let you set effort or write a system prompt, and it sends no effort value; which default the model host applies is not stated on the pages we read.Source: openrouter.ai · checked 2026-11-19 · second-hand
Our table
Fable 5.2 prompting: Anthropic’s general advice with our example prompts
- Anthropic’s Prompting best practices page applies to all current Claude models, Fable 5.2 included. Its general principles are: be clear and direct, add context, use examples, structure prompts with XML tags, give the model a role in the system prompt, and put long documents at the top with the question at the end.Source: platform.claude.com · checked 2026-11-19
Advice for all current Claude models
Fable 5.2 prompting: Anthropic’s general prompting advice for Claude Fable 5.2, paraphrased, with an example prompt of ours
Most of Anthropic’s advice for Claude Fable 5.2 is written for developers who set a system prompt and an effort level through the API. This chat has neither: you write one message at a time. So the example prompts here are single messages. Line breaks become spaces in this chat and angle brackets are removed, so where Anthropic suggests XML tags we use plain labels. On this site a typical reply to Claude Fable 5.2 costs about 11 credits (a credit is about $0.01 in the $5 pack), billed by its real token cost, at least 1 credit.
| Advice | What Anthropic says (paraphrased) | Our example prompt |
|---|---|---|
| Be clear and direct | Claude responds well to explicit instructions. State the output format and constraints, ask outright for anything beyond the bare task, and test the prompt by asking whether a colleague with no context could follow it. | Write a reminder to a tenant who is 5 days late with rent. Tone: polite but firm. Length: under 120 words. Format: a subject line, then two short paragraphs. Also add one line offering a payment plan. |
| Say why | Giving the reason behind an instruction lets the model generalise from it. Anthropic’s example: text that will be read aloud by a speech engine, so no ellipses. | This summary will be read aloud in a meeting, so avoid tables, abbreviations and brackets. Summarise the notes below in six sentences. Notes: [PASTE] |
| Show examples | A few relevant, varied and clearly marked examples improve accuracy and consistency. Anthropic suggests 3 to 5. | Shorten product names to a plain label. Example 1: Ultra Pro Max Wireless Charging Pad 3rd Gen becomes Wireless charger. Example 2: Stainless Steel Insulated Bottle 750ml becomes Water bottle. Example 3: Ergonomic Mesh Office Chair Black becomes Office chair. Now do this one: [NAME] |
| Structure the prompt | XML tags help the model tell instructions, context, examples and inputs apart. Use consistent, descriptive tag names. In this chat angle brackets are removed, so plain capital labels do the same job. | TASK: Compare the two offers. CONTEXT: I run a six-person shop and pay in euros. INPUT: Offer 1 [PASTE] Offer 2 [PASTE]. FORMAT: a table with price, term and risk, then a two-line recommendation. |
| Long text first, question last | For inputs of 20k tokens or more, put the long documents above your question. Anthropic says a question at the end can improve response quality by up to 30 percent in its tests, especially with several documents. | [PASTE THE LONG TEXT] Question about the text above: list every deadline it mentions and who owns each. Answer only from the text. If a deadline has no owner, write no owner. |
Our table
Fable 5.2 prompting tips from Anthropic’s model-specific page, with our example prompts
- Anthropic’s Fable 5.2 prompting page says its numbers come from its own testing on specific tasks, such as replayed coding sessions and small coding tasks. It tells readers to add one piece of guidance at a time and measure its effect and its token cost on their own tasks.Source: platform.claude.com · checked 2026-11-19
Advice specific to Fable 5.2
Anthropic’s Fable 5.2 prompting tips, paraphrased, with an example prompt of ours
These tips come from Anthropic’s page on prompting Fable 5.2 itself. Several are written for agents that run many tool calls, which this chat does not do, so we turn each into a prompt you can send as a message. Anthropic’s own advice is to try one change at a time and measure it. Our table works as a checklist of things to try, and we have not measured any of them. For the council pattern with several models, see our LLM council prompt page.
| Advice | What Anthropic says (paraphrased) | Our example prompt |
|---|---|---|
| Calibrate effort | Effort sets how much the model thinks, how many tool calls it makes, how long a turn takes and how many tokens it uses. Start at high, move down where your tests hold. Effort has little effect on reply length, so use the prompt to set length and tone. This chat has no effort control. | Reply in at most five lines and plain words. Line 1: the answer in one sentence. Lines 2 to 5: the reason. Question: [QUESTION] |
| Do not ask it to think less | On Fable 5.1 a line like "answer directly without deliberating" usually worked. On Fable 5.2 it does not stop thinking at medium and above. Remove such lines and lower the effort instead. | Give the final answer first, then at most three lines on the main reason. Skip the background I already know. |
| Finish the whole task | On long multistep tasks the model sometimes ends a turn early, with a list of what is left or an offer to continue, more often at low and medium effort. Anthropic suggests a persistence paragraph in the system prompt. In its replay test, about half as many sessions ended with unfinished work, and output tokens rose about 30 percent. | Do all five items below in this one reply. If an item cannot be done from what I gave you, say which one and why in the last line, and still finish the rest. Items: 1 [..] 2 [..] 3 [..] 4 [..] 5 [..] |
| Approve specifically | When you tell the model to carry on, say exactly what you approve instead of sending "continue". Vague approval rules can backfire: in one test a rule listing messages to people outside the conversation made the model hold back about half of its routine requests. | Yes, rewrite sections 2 and 3 as you proposed. Do not touch section 1. Keep the table as it is. |
| Say who owns the code | The model pays attention to ownership. A sentence saying it is the owner and long-term maintainer made it consolidate duplicated code in about 75 percent of runs, against under 10 percent without it, but output tokens rose by about 70 to 80 percent and it made changes nobody asked for. Leave the sentence out if the user owns the code and wants minimal changes. | Treat the function below as yours to maintain. Remove the duplication you find and list each change in one line. Do not add tests. [CODE] For a minimal fix, say instead: Change only what causes the bug and nothing else. [CODE] |
| Write for a reader who has not seen the work | People who read a final report usually have not seen the tool calls or files behind it. Anthropic suggests asking for self-contained messages with the key points in the first few sentences, plain words and short sentences. In its tests this made reports less dense. | Write the summary for a reader who has not seen our conversation. Put the decision in the first two sentences, use short plain sentences, and explain any term you made up. |
| Go easy on hard style limits | The model follows style rules closely. A hard limit such as one instruction per sentence in 20 words or fewer can make its prose choppy, so remove rules you no longer need. | Prefer short sentences, but vary their length where it reads better. |
| Ask for conclusions, not its inner reasoning | The safeguards can decline requests that ask the model to reproduce its internal reasoning in the reply (the reasoning_extraction category). Anthropic says to remove such instructions; API users who want to see reasoning read the thinking blocks. | Give your conclusion and the evidence it rests on, in four lines. |
Nearest available
Alternatives to Fable 5.2 prompting you can use now
Claude Fable 5.2
Anthropic$8 / $40 (per million tokens)
- Released on November 9, 2026 (Anthropic’s model page marks it "Latest").Source: platform.claude.com · checked 2026-11-13
- Anthropic describes it as "for demanding reasoning and long-horizon agentic work": hours-long agent tasks, multi-step deep research, and finished documents, spreadsheets and presentations.Source: platform.claude.com · checked 2026-11-13
- Context window 1M tokens; maximum output 128K tokens; text and image input, text output.Source: platform.claude.com · checked 2026-11-13
Claude Opus 5.5
Anthropic$4 / $20 (per million tokens)
- Released on September 22, 2026 (Anthropic’s model page marks it "Latest"); retirement not sooner than September 22, 2027.Source: platform.claude.com · checked 2026-11-15
- Anthropic describes it as "for long-running agentic coding and knowledge work": multihour autonomous coding agents, large-scale refactoring, complex systems engineering, vision-heavy workflows and computer use.Source: platform.claude.com · checked 2026-11-15
- Context window 1M tokens; maximum output 128K tokens; text and image input, text output. Adaptive thinking is always on (default effort "medium"); reliable knowledge cutoff June 2026.Source: platform.claude.com · checked 2026-11-15
Claude Sonnet 5.5
Anthropic$2 / $10 (per million tokens)
- Anthropic’s model page lists Claude Sonnet 5.5 as released on September 28, 2026, status "Active (latest)", with retirement not sooner than September 28, 2027.Source: platform.claude.com · checked 2026-11-17
- Anthropic’s model page says Sonnet 5.5 offers "the best combination of speed and intelligence", built for everyday coding, agent and enterprise workloads. The announcement calls it the second model in the Claude 5.5 family and a faster, lower-cost complement to Claude Opus 5.5, strongest at well-scoped everyday tasks, bug fixing and polished documents, slides and spreadsheets, and says Opus 5.5 remains stronger at complex, open-ended work.Source: anthropic.com · checked 2026-11-17
- Context window 1M tokens, maximum output 128K tokens (up to 300K on the Message Batches API with a beta header), text and image input, text output, adaptive thinking with default effort high on the API, and a reliable knowledge cutoff of June 2026. Setting temperature, top_p or top_k returns a 400 error, and the API does not accept turning thinking fully off (the lowest setting is between_tools).Source: platform.claude.com · checked 2026-11-17
FAQ
Fable 5.2 prompting FAQ
How do I prompt Claude Fable 5.2?
Start with Anthropic’s general advice: be clear and direct, give the reason, add examples, structure the prompt, and put long text before the question. Then add the Fable 5.2 tips one at a time: finish the whole task, approve specifically, say who owns the code, and ask for conclusions rather than inner reasoning. Our tables pair each tip with an example prompt.
Where does this advice come from?
From Anthropic’s documentation, read on November 19, 2026: the "Prompting Claude Fable 5.2" page, the "Prompting best practices" page, the effort page, the Fable 5.2 model page and the launch announcement. Each table row names its source, and the example prompts are our own wording. We have not run them against any model.
What effort should I use with Fable 5.2?
Anthropic lists five levels: low, medium, high, xhigh and max. The API default is high, and it advises starting there and moving down where your own tests hold. Medium is the default in the Claude apps and Claude Code. This site’s chat does not let you set effort, so use the prompt to control length and detail instead.
Does telling Fable 5.2 to answer without thinking work?
Not reliably, according to Anthropic. A line such as "answer directly without deliberating" did not stop the model thinking at medium effort and above, even though it usually worked on Fable 5.1. Anthropic says to remove such lines and lower the effort level, which is an API setting.
Can I try these prompts on this site?
Yes. On this site a typical reply to Claude Fable 5.2 costs about 11 credits (a credit is about $0.01 in the $5 pack), billed by its real token cost, at least 1 credit. A message holds at most 4,000 characters, line breaks become spaces, angle brackets are removed, and you cannot set a system prompt or effort. Our Claude Opus 5.5 page covers a cheaper option, about 6 credits.
Which old prompt habits should I drop?
Anthropic names several: lines telling the model to think less, hard sentence-length limits, blanket "never" rules that pile up in notes, and instructions that make it write out its reasoning. It says Fable 5.2 follows instructions closely, so an outdated rule keeps being obeyed. Re-test prompts you wrote for earlier models.
Is the advice the same for Opus 5.5?
Not exactly. Anthropic gives each model its own prompting page, and says to rerun effort tests rather than carry settings over. For example, its effort page gives medium as the default for Claude Opus 5.5 and high for Claude Fable 5.2. Use the general tips on any Claude model and test the model-specific ones.
Is this the official Claude website?
No. This page is unofficial and not affiliated with or endorsed by Anthropic. Facts link to the pages we read.