# MagicClaude Full LLM Guide MagicClaude is a source-backed gallery of impressive Claude demos with reproducible prompts. The site is scoped to "wow demos with prompts" and intentionally links adjacent Claude topics to sibling network sites instead of duplicating them. Important disclaimer: MagicClaude is independent and not affiliated with Anthropic. Last model refresh: 2026-09-20. ## Editorial Rules 1. Capability claims should cite official Anthropic documentation or Claude Help Center pages. 2. Prompts are demo scripts, not benchmarks. MagicClaude publishes no scores and runs no evaluations of its own. 3. Output can vary by account, model, plan, settings, date, and product surface. 4. High-risk workflows need guardrails and human verification. 5. The site should preserve wonder while making setup and verification visible. 6. Every demo names the current Claude model it is cast on. A superseded model name is treated as a defect, not a dated detail. 7. No page reproduces a Claude transcript. Pages describe what a prompt should produce and how to check it. ## Current Model Lineup Per Anthropic's model overview (https://docs.anthropic.com/en/docs/about-claude/models) and choosing-a-model guide (https://platform.claude.com/docs/en/about-claude/models/choosing-a-model), the generally available models as of September 2026 are: - Claude Fable 5.1, API ID `claude-fable-5-1`. Generally available 2026-09-01 and described by Anthropic as its most capable generally available model. 1M-token context, 128K max output, always-on adaptive thinking, default effort high in Claude Code and medium on Claude.ai and Cowork. MagicClaude casts it for long-horizon agentic demos. - Claude Opus 5, API ID `claude-opus-5`. Generally available 2026-07-24. 1M-token context, 128K max output, adaptive thinking, optional fast mode. Default on Claude Max and strongest model on Claude Pro. This is the default cast for most MagicClaude demos. - Claude Sonnet 5, API ID `claude-sonnet-5`. 1M-token context, 128K max output. Cast for rehearsal and latency-sensitive runs. - Claude Haiku 4.5, API ID `claude-haiku-4-5`. 200K-token context, 64K max output. Cast for cheap dry runs. Anthropic guidance is to start with Opus 5 and move to Fable 5.1 for demanding reasoning and long-horizon agentic work. Claude Opus 4.8 is no longer the current Opus and its source record was retired from this site on 2026-09-19. MagicClaude names models so prompts stay reproducible. It does not rank them. Scored comparisons, benchmark methodology, and pricing analysis belong to https://claudereports.com; general model selection belongs to https://claudecentral.com. ## Pages ### Homepage URL: https://magicclaude.com/ Introduces MagicClaude, links to The Stage, the launch demo cabinet, explains the difference between prompt lists and demo scripts, and links to primary-source examples. ### The Stage URL: https://magicclaude.com/stage/ Signature experience. Spotlights one sourced wow-demo: the one-prompt artifact prototype on Claude Opus 5 (`claude-opus-5`). Includes the copyable prompt, house/proof cue checklists, verification steps, and Anthropic artifact citations. States that output varies and does not publish a Claude transcript. Funny fails route to https://claudegonewild.com; experiments route to https://curiousclaude.com. The long-horizon Fable 5.1 demo is named as the understudy, not a second spotlight. Casting details remain on /casting/. ### Demo Gallery URL: https://magicclaude.com/demos/ Lists all demos: artifact prototype, PDF evidence table, screenshot UX audit, CSV data visualizer, codebase debugging, browser computer use, web research brief, long-context synthesis, long-horizon agent run, and effort dial. Each card names the model the demo is cast on. ### Casting URL: https://magicclaude.com/casting/ Says which current Claude model to run each demo on, with API IDs, context limits, and the cases where a model is the wrong cast. Most demos are cast on Claude Opus 5; the long-horizon agent run is cast on Claude Fable 5.1 and the effort dial uses both. The page explicitly routes model-selection guidance to Claude Central and scored comparisons to Claude Reports. ### Free Tools URL: https://magicclaude.com/tools/ Index of free client-side tools for finding, remixing, and rating Claude demos. The tools do not call Claude automatically, do not require a signup, and do not send user inputs to a backend. ### Demo Gallery Tool URL: https://magicclaude.com/tools/demo-gallery/ Filterable gallery of exact Claude demo prompts. Each card includes output type, audience, difficulty, expected result, copyable prompt text, and verification steps. It targets queries such as "impressive Claude demos" and "Claude prompt examples." ### Prompt Remixer URL: https://magicclaude.com/tools/prompt-remixer/ Browser-only prompt generator that starts from MagicClaude-style demo patterns and lets the reader change theme, audience, reveal style, output format, and guardrails. It preserves assumptions, expected reveal, and verification instructions. ### Wow Meter Checklist URL: https://magicclaude.com/tools/demo-rater/ Interactive scoring checklist for Claude demo results. It scores first-try reliability, prompt fidelity, surprise, usefulness, verification quality, and safety/data hygiene, then generates a copyable freshness submission format. ### Artifact Prototype URL: https://magicclaude.com/demos/artifact-prototype/ Demonstrates how to ask Claude for a shareable interactive artifact from a product idea. Primary sources include Claude Help Center pages for artifacts, publishing artifacts, and custom visuals. ### PDF Evidence Table URL: https://magicclaude.com/demos/pdf-evidence-table/ Demonstrates PDF analysis as an evidence-clerk workflow. Primary sources include Anthropic PDF support, citations, structured output, and prompting docs. ### Vision UI Audit URL: https://magicclaude.com/demos/vision-ui-audit/ Demonstrates screenshot analysis for UX and accessibility observations. Primary sources include Anthropic vision docs, prompting guidance, structured output guidance, and custom visual notes. ### Data Visualizer URL: https://magicclaude.com/demos/data-visualizer/ Demonstrates CSV-style analysis and interactive visual generation. Primary sources include code execution, custom visuals, artifacts, and structured output guidance. ### Codebase Debugging URL: https://magicclaude.com/demos/codebase-debugging/ Demonstrates a reproduce-fix-verify loop in Claude Code. Primary sources include Claude Code overview, text editor tool, code execution tool, and prompting guidance. ### Browser Computer Use URL: https://magicclaude.com/demos/browser-computer-use/ Demonstrates safe, non-sensitive browser or computer-use workflows. Primary sources include Anthropic computer-use safety guidance and web search docs. ### Web Research Brief URL: https://magicclaude.com/demos/web-research-brief/ Demonstrates current-source research with confidence labels and unknowns. Primary sources include Claude web search, API web search, Research, and citations docs. ### Long-Context Synthesis URL: https://magicclaude.com/demos/long-context-synthesis/ Demonstrates structured synthesis of long transcripts or document bundles. Primary sources include context-window docs and the current Fable 5.1, Opus 5, and Sonnet 5 model docs, which each carry a 1M-token context window. ### Long-Horizon Agent Run URL: https://magicclaude.com/demos/long-horizon-agent-run/ Demonstrates a supervised long-running agentic task on Claude Fable 5.1: written acceptance criteria, a checkpointed PLAN.md the model updates per item, per-item verification commands, commits per item, and a final report that separates verified from unverified work. Primary sources include the Fable 5.1 overview, the Fable 5.1 announcement, the choosing-a-model guide, Claude Code docs, and code execution docs. It is explicitly not a benchmark of agentic ability. ### Effort Dial URL: https://magicclaude.com/demos/effort-dial/ Demonstrates thinking-effort behavior by running one identical prompt at low and high effort and diffing what each answer questioned. Primary sources include the Fable 5.1 overview, which documents default effort levels per surface, and the Opus 5 overview. The page states that two runs of one prompt prove nothing general and routes scored comparisons to Claude Reports. ### Prompt Patterns URL: https://magicclaude.com/prompts/ Explains the reusable MagicClaude prompt scaffold: role, goal, context, requirements, and source boundary. ### Claude Demo Prompts Guide URL: https://magicclaude.com/guides/claude-demo-prompts/ High-intent guide for choosing Claude prompt examples by proof shape: clickable artifact, evidence table, visual critique, chart, source-backed research brief, or long-context synthesis. ### Show Claude to a Team URL: https://magicclaude.com/guides/show-claude-to-a-team/ Run-of-show guide for presenting Claude to a team without overclaiming. It recommends one anchor demo, one backup demo, a visible verification checklist, and an explicit anti-claim. ### Freshness URL: https://magicclaude.com/freshness/ Weekly-extensible update surface for new demos, source refreshes, and retired claims. ### Sources URL: https://magicclaude.com/sources/ Index of official sources cited by the site. ### About URL: https://magicclaude.com/about/ Methodology, independent disclaimer, and Claude Network links. ## Claude Network Routing - Broad ecosystem hub and model-selection guidance: https://claudecentral.com - Weekly news: https://claudeweekly.com - Benchmarks, scored model comparisons, and pricing analysis: https://claudereports.com - Claude Code engineering practice: https://claudeshipscode.com - Claude skills: https://claudehasskills.com - Context engineering: https://claudecontext.com - Experiments and edge cases: https://curiousclaude.com - Funny, weird, or slapstick outputs: https://claudegonewild.com ## Primary Source Themes - Artifacts: shareable apps, tools, documents, code, diagrams, HTML, and React components. - Publishing artifacts: public artifact viewing/interacting and embed controls. - Claude Code: codebase reading, file editing, command execution, and developer-tool integration. - Code execution: sandboxed Python/Bash for analysis and file generation. - Vision: image inputs through Claude.ai, Workbench, and API image blocks. - PDF support: text, pictures, charts, and tables in PDFs. - Web search and Research: current information with citations and multi-search investigation. - Computer use: permissioned desktop/browser workflows with safety limits. - Prompting and structured outputs: explicit output formats and schema conformance when needed. - Context/model docs: context-window management and current model capabilities. - Model lineup: generally available models, API IDs, context limits, and retirement dates. - Thinking effort: adaptive thinking and per-surface default effort levels on Fable 5.1 and Opus 5. ## Free Tool Summary - Demo Gallery Tool: filter exact prompt examples and copy a reproducible script. - Prompt Remixer: create a themed prompt variant while preserving the verification loop. - Wow Meter Checklist: score and format a demo result for future freshness review. All tools run locally in the browser and should be treated as prompt-planning aids, not as Claude API clients.