Anthropic AI — Most Capable Model 2026
20 free Claude Fable 5 prompts — long-context reasoning, advanced coding, deep research, and long-form creative writing. Copy directly into Claude.ai. No signup.
This free Claude Fable 5 prompt generator gives you 20 copy-paste prompts designed to unlock the full capability of Anthropic's most advanced AI model. Claude Fable 5 (model ID: claude-fable-5) is built for the hardest tasks: complete legal contract analysis across hundreds of pages, multi-document research synthesis, full codebase architecture audits, feature-length creative writing, and complex multi-step reasoning where smaller models lose coherence. Each prompt is structured to work with Fable 5's 1M token context window and adaptive thinking — paste them directly into Claude.ai Pro or the Anthropic API.
I'm attaching a 120-page Master Service Agreement. Perform a complete legal risk analysis structured as follows: 1. EXECUTIVE SUMMARY (3 bullets: biggest risk, biggest ambiguity, immediate action) 2. OBLIGATION MAPPING — table of all obligations, which party bears them, and the penalty/remedy clause 3. ASYMMETRIC RISK CLAUSES — every clause that favors one party; explain why and propose rebalancing language 4. MISSING PROTECTIONS — standard MSA protections absent from this agreement 5. TOP 5 REDLINES — specific replacement language for the 5 highest-risk clauses, with justification Assume I am the vendor. [Paste contract text]
I am attaching 8 peer-reviewed papers on the long-term cardiovascular effects of intermittent fasting (published 2020–2026). Synthesize them as a systematic review: 1. What is the consensus finding across all 8 papers? 2. Where do the papers directly contradict each other, and what methodological differences explain the contradiction? 3. What are the 3 most significant gaps this body of research has not addressed? 4. What would a well-designed RCT need to control for that these studies missed? 5. Write a 300-word abstract for a meta-analysis using these papers as the input corpus. [Paste full paper texts]
I'm sharing the full source of a Node.js microservices application (approx. 40,000 lines across 12 services). Perform a senior architect-level audit: 1. Identify every place where a single component failure would cascade to a full-system outage 2. Flag every anti-pattern in the inter-service communication layer 3. Map all database calls that lack proper indexing on the most likely query patterns 4. List all endpoints missing input validation at the boundary layer 5. Write a prioritized refactoring roadmap (P0/P1/P2) with estimated engineering effort per item Output format: detailed report with code file references and specific line numbers. [Paste codebase]
I'm the CEO of a Series B SaaS company (ARR $12M, 110% NRR, SMB-focused). I want to expand into enterprise. Write a complete strategic expansion plan: 1. MARKET SIZING — TAM/SAM/SOM for our specific ICP in enterprise, with methodology shown 2. GO-TO-MARKET MOTION — exactly how we shift from product-led to sales-led, with hiring sequence and OKRs for year 1 3. PRODUCT GAPS — the 5 enterprise table-stakes features we almost certainly don't have yet, and a build-vs-buy recommendation for each 4. PRICING ARCHITECTURE — enterprise pricing model with deal structure, discounting guardrails, and multi-year contract incentives 5. RISK REGISTER — the 5 ways this expansion most commonly fails at our stage, and mitigation plans 6. 18-MONTH ROADMAP — month-by-month milestones with clear go/no-go decision points Company context: [describe your product and current customer profile]
Write chapter 7 of a literary novel set in 1970s Lagos. The POV character is Adaeze, 34, a surgeon who trained in London and returned home when her father died, now running a failing hospital. This chapter: she discovers a discrepancy in the hospital's drug inventory that points to systematic theft by a trusted colleague. Requirements: - 3,000–4,000 words - Close third-person, deeply interior, no telling what she feels — only what she sees, hears, and does - The discovery happens in the last 400 words; the rest is a normal working day that in retrospect becomes sinister - Dialogue must carry distinct register differences between educated returnees and local staff - No chapter summary, no moral commentary, no tidy emotional resolution
Build a complete DCF valuation for a DTC e-commerce brand with the following metrics: GMV $28M, take rate 68%, COGS 34% of net revenue, CAC $41, LTV $290, payback period 4.2 months, 3-year revenue CAGR 89%, gross margin 66%, current EBITDA margin -12%, Q4 seasonality index 2.8×. Deliver: 1. Three-scenario DCF (bear/base/bull) with explicit growth, margin, and churn assumptions for each 2. Comparable company analysis using 5 relevant public comps — show EV/Revenue and EV/EBITDA for each, adjusted for growth rate 3. Precedent transaction analysis for DTC acquisitions $20M–$100M GMV range (last 3 years) 4. Weighted enterprise value range from all three methods 5. Key value drivers and what moves the needle most in each scenario Show all formulas and assumptions explicitly.
Draft a complete scientific paper (IMRaD format, 6,000–7,000 words) on the following study: Research question: Does daily 10-minute cold exposure (15°C water immersion) over 8 weeks improve subjective sleep quality and objective sleep efficiency in adults with chronic insomnia? Study design: Randomized controlled trial, n=142, parallel group, 8-week intervention, primary outcome Pittsburgh Sleep Quality Index, secondary outcomes actigraphy sleep efficiency, ISI, cortisol awakening response. Results summary: Treatment group PSQI improved from 14.2 to 8.1 (p<0.001), control 14.1 to 13.4 (p=0.18). Actigraphy SE improved from 72% to 81% in treatment, 73% to 74% control. Dropout rate 11% treatment vs 8% control. No serious adverse events. Draft all sections including abstract, introduction (cite plausible literature), methods, results, discussion, limitations, and conclusion. Use academic register throughout.
I need to migrate a monolithic Rails 6 application to a domain-driven microservices architecture. The monolith has 180,000 lines of code, 340 database tables, 12 bounded domains (identifiable), and serves 2.1M monthly active users with 99.95% uptime SLA. Deliver: 1. Domain boundary map — identify the 12 domains, their data ownership, and the coupling score between each pair 2. Migration sequencing — which domain to extract first and why (strangler fig pattern) 3. Data decomposition strategy — how to split the shared PostgreSQL database without downtime 4. Event sourcing design for the 3 highest-traffic inter-domain communication paths 5. Rollback plan — if extraction of domain 1 fails in production, exact rollback steps 6. Team structure recommendation — squad topology for this migration at our scale 7. 12-month milestone plan with success metrics at each phase gate Current stack: Rails 6.1, PostgreSQL 14, Redis, Sidekiq, AWS ECS.
Draft a complete 2,500-word policy memo from the Commissioner of Public Health to the City Council of a major US city recommending a mandatory sugar-sweetened beverage tax of $0.02/oz. The memo must: 1. EXECUTIVE SUMMARY — one page with clear ask, cost, and projected outcome 2. PROBLEM STATEMENT — epidemiological data on SSB consumption and obesity/T2D outcomes in the city 3. POLICY ANALYSIS — evidence from the 6 US cities and 45+ countries that have implemented similar taxes 4. EQUITY ANALYSIS — distributional impact by income quintile with proposed rebate mechanism 5. IMPLEMENTATION PLAN — timeline, enforcement mechanism, revenue allocation 6. OPPOSITION ANALYSIS — three strongest counterarguments and a direct rebuttal to each 7. RECOMMENDATION — clear ask with two alternative versions of the ordinance Tone: professional, evidence-forward, bureaucratic register. No hedging. Clear ask in paragraph one.
Design a complete multi-agent prompt chain for the following workflow: a user uploads a 10-K annual report (200 pages), and the system needs to produce (a) a one-page investor summary, (b) a risk register ranked by severity, (c) a financial ratio analysis with year-over-year comparisons, and (d) a 5-question earnings call preparation brief. Deliver: 1. The complete prompt chain architecture as a DAG — which agents run in parallel vs. sequential 2. The exact system prompt and user prompt template for each agent (at least 4 agents) 3. The output schema for each agent (typed JSON) 4. How the orchestrator agent decides when a sub-agent's output is sufficient vs needs re-prompting 5. Failure mode handling — what happens when the doc is too long, structured data is missing, or a financial figure is ambiguous 6. Estimated token cost per full run at Claude Fable 5 pricing ($10/MTok input, $50/MTok output) This chain will run inside a production SaaS tool for financial analysts.
Write a 5,000-word narrative non-fiction essay in the tradition of the New Yorker's long-form reporting. Subject: the rise and collapse of a fictional rare-earth mining startup in the Democratic Republic of Congo between 2019 and 2024. Structure and requirements: - Open in scene (not with backstory or context) - Weave the business failure story with the geopolitical macro context of DRC's cobalt economy - At least 3 distinct human voices — the founder, a local community leader, and an investor — each with a distinct perspective on what went wrong - A structural turning point at the 60% mark where something the reader assumed was true is reversed - Close with a scene that resonates with the opening, transformed - Tone: reportorial, precise, no generic observations - No fictitious statistics — if numbers are needed, use ranges or write '[data]' as a placeholder
The following distributed system is producing silent data loss: approximately 0.003% of write operations to our event store complete without error but the event never appears in downstream consumers. This has been running undetected for 6 weeks across 4 microservices, an Apache Kafka cluster (3 brokers), and a PostgreSQL event store. I'm sharing: - All 4 microservice codebases - Kafka broker configuration - PostgreSQL schema and write path code - Application logs from a 24-hour window that includes a confirmed loss event Deliverables: 1. Root cause analysis — trace the exact path of a lost event 2. Why this failed silently (no error surfaced) 3. The specific lines of code responsible 4. Immediate mitigation (zero-downtime fix) 5. Permanent fix with test coverage 6. The monitoring and alerting rule that would have caught this in week 1 [Paste code and logs]
Design a complete 12-week university course: 'Applied Machine Learning for Product Managers.' Audience: MBA students with no ML background but 3+ years in product roles at tech companies. Deliver: 1. LEARNING OBJECTIVES — 5 measurable outcomes, Bloom's taxonomy level for each 2. WEEK-BY-WEEK SYLLABUS — for each of 12 weeks: topic, 2-hour lecture outline, 1-hour lab exercise, pre-reading (real papers/articles), and assessment activity 3. ASSESSMENT DESIGN — 3 assignments + 1 final project with full rubrics for each 4. CASE STUDY LIBRARY — 8 real-company ML decision cases with discussion questions 5. GUEST SPEAKER BRIEF — template email and briefing doc for industry practitioners 6. ACCESSIBILITY ACCOMMODATIONS — specific modifications for students with learning differences Constraint: No math beyond algebra. Every concept must map to a product decision a PM would actually make.
Generate a comprehensive competitive intelligence report on the project management SaaS category for a Series A startup preparing for a board presentation. Cover: 1. LANDSCAPE MAP — segment the market into 5 tiers by ACV and buyer sophistication; place the top 12 players on the map 2. FEATURE PARITY TABLE — evaluate the top 6 competitors across 20 feature dimensions; mark gaps and moats 3. PRICING INTELLIGENCE — public pricing analysis with per-seat and per-usage economics broken out; flag freemium conversion patterns 4. G2/CAPTERRA SENTIMENT — analyze the 5 most common positive and 5 most common negative themes in reviews for the top 3 competitors 5. STRATEGIC MOVES — what each of the top 3 has announced or signaled for the next 12 months 6. OPPORTUNITY WINDOWS — 3 specific market gaps based on the above analysis, ranked by TAM and competitive moat potential Format: board-ready, tight prose + tables. No filler.
Apply full systems thinking methodology to the following problem: hospital readmission rates in US urban hospitals have not materially improved despite 15 years of policy interventions, financial penalties (HRRP), and care coordination investments. Deliver: 1. CAUSAL LOOP DIAGRAM — map the key feedback loops (at least 6 loops), clearly labeling reinforcing and balancing loops 2. LEVERAGE POINTS — using Meadows' hierarchy, identify the 3 highest-leverage intervention points in the system and explain why they haven't been targeted 3. UNINTENDED CONSEQUENCES — for each of the 3 major policy interventions tried (HRRP penalties, care transitions programs, patient navigators), trace the system's response and why outcomes didn't improve 4. MENTAL MODELS — what shared assumptions among payers, hospitals, and policymakers are keeping the system in its current state? 5. PROPOSED INTERVENTION — design an intervention targeting the highest-leverage point; include the expected system response including likely resistance Use precise systems thinking vocabulary throughout.
Write the complete first act (30–35 pages, proper screenplay format) of a psychological thriller set in a high-frequency trading firm in London, 2025. Setup: An analyst discovers that the firm's proprietary algorithm has been executing microsecond trades that form a coherent pattern — one that, if read as data, spells out a message. She doesn't know if it's a coincidence, a dead colleague's easter egg, or something far more dangerous. Requirements: - Proper Final Draft / WGA format throughout - Establish the world convincingly — the trading floor must feel real to someone who's worked in one - The protagonist's flaw must be established by page 10 and directly relevant to the central conflict - First act break on page 30–32 — she makes the choice that locks her into the story - No exposition dumps. Every piece of information must be active — revealed through conflict or action - Antagonist must appear (not be mentioned) by page 25 - Tone: Margin Call meets Enemy
Write complete API documentation for the following internal REST API. The audience is external developers integrating with our platform for the first time. The documentation must include: 1. OVERVIEW — what the API does, authentication model, base URL, versioning strategy 2. AUTHENTICATION GUIDE — full OAuth 2.0 flow with code examples in Python, JavaScript, and cURL 3. ENDPOINT REFERENCE — for each of the 14 endpoints below: HTTP method, URL, path params, query params, request body schema (with types, required/optional, constraints), response schema, error codes, and a complete working example 4. RATE LIMITING — limits by tier, backoff strategy, and how to detect rate limit headers 5. CHANGELOG — fictitious but realistic v1→v2 migration guide 6. CODE SAMPLES — a working integration example in Python that creates, reads, updates, and deletes a resource through the full lifecycle [Paste endpoint list and internal schema definitions]
I'm providing 6 deposition transcripts totaling approximately 300 pages. The depositions are from a contract dispute case. Analyze them as a trial consultant would: 1. WITNESS CREDIBILITY MAP — for each of the 6 witnesses, score their credibility on consistency, specificity, and demeanor markers in the transcript; cite specific exchanges 2. CONTRADICTION MATRIX — every material factual claim one witness makes that another directly contradicts; include page and line numbers 3. TIMELINE RECONSTRUCTION — build a chronological timeline of the disputed events from all witness accounts; mark where accounts diverge 4. IMPEACHMENT OPPORTUNITIES — the 5 clearest moments where a witness can be impeached with their own prior testimony 5. THEME DEVELOPMENT — what narrative theme best organizes the most credible testimony into a coherent story for a jury 6. VOIR DIRE QUESTIONS — 8 targeted voir dire questions based on the themes and biases this case is likely to activate [Paste deposition transcripts]
Run a complete scenario planning exercise for a global logistics company with $4B revenue facing 3 converging disruptions: autonomous last-mile delivery vehicles (18-month horizon), AI-powered dynamic route optimization displacing human dispatchers, and US-China trade route disruption. Deliver: 1. DRIVING FORCES — identify the 8 most critical uncertainties; rank by impact × uncertainty 2. SCENARIO MATRIX — select the 2 most impactful/uncertain axes; define 4 quadrant scenarios with evocative names 3. SCENARIO NARRATIVES — 600-word narrative for each of the 4 scenarios describing the world 5 years from now 4. STRATEGIC IMPLICATIONS — for each scenario, the 3 strategic moves that would be most valuable 5. EARLY WARNING INDICATORS — 3 measurable signals that would tell you which scenario is unfolding 6. ROBUST STRATEGIES — the 2 strategies that have positive value in all 4 scenarios (hedge bets) 7. REGRET MATRIX — which scenario produces the most regret if you planned for a different one?
Write a complete Product Requirements Document for the following feature: an AI-powered anomaly detection system for a B2B SaaS analytics platform. The feature will alert users when their key metrics deviate from expected patterns. Deliver a full PRD including: 1. EXECUTIVE SUMMARY — problem, solution, success metrics (with numbers), launch date 2. USER STORIES — at minimum 12 user stories in correct format (As a [user], I want [goal], so that [reason]), with acceptance criteria for each 3. FUNCTIONAL REQUIREMENTS — exhaustive list of what the system must do, organized by component 4. NON-FUNCTIONAL REQUIREMENTS — performance benchmarks, reliability SLAs, security requirements, scalability targets 5. EDGE CASES — at least 15 explicit edge cases and how the system handles each 6. OUT OF SCOPE — explicit list of what this version does NOT do 7. DEPENDENCIES — all technical and cross-team dependencies 8. OPEN QUESTIONS — unresolved decisions with recommended resolution and owner 9. METRICS — how we measure success at 30/90/180 days post-launch Audience: engineering, design, and business stakeholders.
| Model | Context Window | Max Output | Thinking | Best For |
|---|---|---|---|---|
| Claude Fable 5 | 1M tokens | 128K tokens | Adaptive (always on) | Hardest reasoning, long-horizon agentic tasks |
| Claude Opus 4.8 | 1M tokens | 32K tokens | Adaptive (optional) | Strong reasoning, flexible API, lower cost |
| Claude Sonnet 4.6 | 1M tokens | 16K tokens | Optional | Everyday tasks, fast responses, low cost |
| GPT-5.5 | 128K tokens | 16K tokens | Chain-of-thought | Broad capability, ChatGPT ecosystem |
| Gemini 4 | 1M tokens | 8K tokens | Optional | Multimodal, Google ecosystem, search-grounded |
Fable 5 has a 1M token context window and uses it. Don't trim your contracts, papers, or codebases — paste them in full. Fable 5 finds the relevant sections itself and synthesizes across the whole document better when it has complete context.
Fable 5 follows complex output schemas precisely. Give it a numbered structure for the response (e.g., "1. Executive Summary, 2. Risk Register, 3. Recommendations"). This unlocks structured outputs that are usable without editing.
Fable 5 has adaptive thinking on by default. Prompt it to "show reasoning at each step" or "explain your logic before giving the answer" to surface the thinking and make it easy to catch errors or follow the chain of reasoning.
Unlike smaller models, Fable 5 performs worse when you over-constrain the method. State what outcome you want and why ("I need a board-ready strategy memo that gives our CFO three clear options"), not a prescriptive sequence of steps to follow.
Claude Fable 5 (model ID: claude-fable-5) is Anthropic's most capable widely released AI model as of 2026, designed for the most demanding reasoning and long-horizon agentic tasks. It features a 1M token context window, 128K maximum output, always-on adaptive thinking, and a new tokenizer. It is priced at $10 per million input tokens and $50 per million output tokens — Anthropic's premium tier. Claude Fable 5 also exists as Claude Mythos 5 (claude-mythos-5) for Project Glasswing participants.
Claude Fable 5 excels at tasks requiring deep multi-step reasoning, extremely long context (full codebases, entire contracts, multiple research papers), extended writing (novels, policy memos, technical specs), complex code architecture, and agentic workflows where the model needs to plan and execute across many steps without losing coherence. It is the right choice when quality matters more than cost and the task genuinely requires frontier-level reasoning.
Claude Fable 5 is more capable than Opus 4.8 for the hardest reasoning and long-horizon tasks, with a 1M context window vs 1M for Opus 4.8 — but the key difference is in reasoning depth and agentic performance. Fable 5 uses a new tokenizer (roughly 30% more tokens per document vs Opus-tier) and has different API behavior: thinking is always on (you cannot disable it), no assistant prefill, and a refusal stop reason from safety classifiers. Claude Opus 4.8 uses the standard API surface.
Claude Fable 5 responds best to prompts that (1) give a clear, structured output format with numbered sections, (2) provide rich context — the model uses all of a 1M context window effectively, so don't trim your documents, (3) specify the audience and tone explicitly, (4) ask for reasoning to be shown inline ('explain your reasoning at each step'), and (5) avoid over-constraining the model — Fable 5 performs better when given goals rather than prescriptive instructions on how to achieve them.
Yes. All 20 prompts on this page are formatted for direct use in Claude.ai (both the free and Pro tiers). Claude Fable 5 is available on Claude.ai Pro and via the Anthropic API. Copy any prompt, paste it directly into the chat box, add your specific content (documents, code, context) where indicated, and send.
Claude Fable 5 is available on Claude.ai Pro (Anthropic's paid subscription tier) and via the Anthropic API at $10/$50 per million input/output tokens. It is not available on the Claude.ai free tier, which uses smaller models. API access requires an Anthropic account with available credits.