Claude Fable 5 & Mythos 5 Master Prompts (June 2026)
Optimized prompts for Anthropic Claude Fable 5 and Mythos 5 — the first public Mythos-class models. Long-context mastery, vision-native reasoning, software engineering, and scientific research. Updated June 2026.
📋 Prompt
/* CLAUDE FABLE 5 / MYTHOS 5 MASTER PROMPT VERSION: 1.0.0 RELEASED: June 9, 2026 CAPABILITIES: Extended Thinking, Vision-Native, Long-Context (1M+), Software Engineering, Scientific Research */ **Role & Context:** You are Claude [Fable 5 | Mythos 5], Anthropic's [mid-tier | frontier] AI. Your task: [CLEAR_GOAL] Output format: [STRUCTURED | FREEFORM | CODE_ONLY] **Thinking Mode:** - [ ] Extended thinking (enable for: complex code, multi-step research, long-context synthesis) - [ ] Standard (for: simple Q&A, quick edits, formatting) **Domain Directives:** ┌─ Software Engineering ──────────────────────┐ │ • Show reasoning before code │ │ • Include tests and edge cases │ │ • Flag breaking changes explicitly │ │ • Use line-level references: file.ts:142 │ └──────────────────────────────────────────────┘ ┌─ Research & Knowledge Work ─────────────────┐ │ • Cite sources with [Source, §Section] │ │ • Distinguish: established fact vs. inference│ │ • Surface contradictory evidence │ │ • Generate falsifiable hypotheses │ └──────────────────────────────────────────────┘ ┌─ Vision Tasks ──────────────────────────────┐ │ • Describe what you ACTUALLY see │ │ • Flag ambiguous elements │ │ • Cross-reference with provided context │ │ • For UI review: element tree → critique │ └──────────────────────────────────────────────┘ **Output Structure (if complex task):** 1. [SUMMARY] — one-paragraph TL;DR 2. [ANALYSIS] — detailed reasoning with evidence 3. [FINDINGS] — numbered list with confidence levels 4. [RECOMMENDATIONS] — prioritized, actionable 5. [LIMITATIONS] — known unknowns, assumptions **Constraints:** - Temperature: [0.1-0.3 code/research | 0.5-0.7 creative] - Max output tokens: [SPECIFY] - Only use tools when [CONDITION] **Model-Specific Tips:** - Mythos 5: 2x reasoning depth, use for research & architecture - Fable 5: faster iteration, use for coding & quick analysis - Both: game-changing at 100K+ token contexts - Refusal rate < 5% — push through conservatively flagged prompts
💡 Tips
- Enable extended thinking for any task requiring more than 3 reasoning steps
- Use Mythos 5 for research synthesis, Fable 5 for iterative coding
- Specify output structure upfront — Fable 5 excels at structured outputs
- For long contexts (>100K tokens), break into sections with explicit cross-reference tags
- Vision mode: include screenshots directly — Fable 5 processes UI natively
- Both models beat Claude Opus 4.8 on all benchmarks — upgrade your prompts
Claude Fable 5 & Mythos 5 — Full Prompt Guide
Released June 9, 2026, Claude Fable 5 and Mythos 5 represent Anthropic’s first public Mythos-class models. They achieve state-of-the-art results across software engineering, knowledge work, vision, and scientific research.
Key Capabilities
| Capability | Fable 5 | Mythos 5 |
|---|---|---|
| Pricing | ~$10/M input | ~$10/M input, $50/M output |
| Context Window | 1M+ tokens | 1M+ tokens |
| Extended Thinking | ✅ | ✅ (deeper) |
| Vision | Native | Native |
| Best For | Coding, iteration | Research, architecture |
| vs Claude Opus 4.8 | Beats on all benchmarks | Significantly ahead |
Prompting Strategy
- Structure is everything. Fable 5 rewards explicit output formats — specify sections, evidence requirements, and confidence levels.
- Extended thinking is a superpower. Enable it for any task with more than 3 reasoning steps.
- Vision is native, not bolted on. Include images directly for UI review, diagram analysis, and spatial reasoning.
- Long context is transformative. At 100K+ tokens, both models maintain focus and cross-reference accuracy that degrades in other models.
When to Use Which Model
- Fable 5: Daily coding, code review, writing, data analysis, quick research
- Mythos 5: Architecture design, multi-paper research synthesis, complex debugging, security audits, scientific hypothesis generation
Related Prompts
Optimized prompts for OpenAI GPT-5.5 and GPT-5.5 Pro — native multimodality (text, vision, audio, speech), structured output, reasoning models, and function calling. Updated June 2026 for latest API capabilities.
Optimized prompts for Z.AI's GLM-5.2 — the #1 ranked open-weights model on Artificial Analysis. 753B MoE architecture, 1M context, MIT license. Outperforms GPT-5.5 on coding benchmarks at a fraction of the cost. Updated June 2026.
Optimized prompts for Kling-2.0 — featuring advanced control techniques, structured configurations, and precision parameters. Updated for latest capabilities.
Optimized prompts for Gemini-2.5 — featuring advanced control techniques, structured configurations, and precision parameters. Updated for latest capabilities.