Frontier Models▸What are people actually building with the newest frontier models?▸time machine

What are people actually building with the newest frontier models?

version 13 · published 6d ago · superseded by v36

Verdict

People are building revenue-generating applications with frontier models: automated trading bots ^601,609,102, legal entity setup ¹⁰¹, faceless YouTube channels ^92,467, AI-generated ad creation ^90,210,599, and an OnlyFans persona earning $43,000 in 30 days ¹⁶⁶. The dominant pattern is pairing a strong planner (e.g., Opus 4.8) with a fast executor (e.g., Fable 5) for cost-performance gains ^88,351. Fable 5 is the most cited model, used for complex multi-step tasks ^95,538, autonomous self-improvement ⁹⁸, and rapid game development ^164,311,439. Newer practices include using Gemini models for voxel heart animation ²¹⁴, structured JSON prompts for video generation ²⁰⁹, multi-agent Etsy store management ³¹², and local AI servers replacing cloud subscriptions ^96,250,505. Many claims remain emerging with single-source corroboration.[24]

Corroborated practices

For Fable 5, using /loop for autonomous multi-step work and giving objectives instead of step-by-step instructions yields better results (corroborated by 5 sources) ⁸⁰. Fable 5 can autonomously self-improve and outperform Opus 4.8 on complex tasks at lower cost (corroborated by 6 sources) ⁹⁸. Fable 5 enables rapid full-game development from idea to code in about one hour (corroborated by 3 sources) ¹⁶⁴. A solo operator built an AI-generated OnlyFans persona using Claude 4.8 Opus, earning $43,000 in 30 days with $400 monthly compute cost (corroborated by 2 sources) ¹⁶⁶. Combining multiple frontier models can automate ad creation from product photos to video (corroborated by 3 sources) ⁹⁰. Using Fable 5 and ClaudeKit's EngineerKit, a single slash command can generate a professional website without a designer or developer (corroborated by 6 sources) ¹⁷¹. Using native Obsidian skills instead of generic HTML parsing reduces context usage by over 99% for AI agents searching vaults (corroborated by 3 sources) ¹⁷⁰. A leaked system prompt can recreate a model's personality on a different base model (corroborated by 2 sources) ⁸⁵. Fable 5 can produce high-quality, animated websites by combining reference screenshots, pre-built components, and specific animation tools with iterative prompting (corroborated by 4 sources) ²⁵⁴. Fable 5 can transform a folder of notes into a connected knowledge graph that answers in your own words and accelerates learning (corroborated by 2 sources) ²⁵². Running Claude Code on an e-ink tablet (reMarkable Paper Pro) creates a distraction-free, eye-comfortable coding environment (corroborated by 2 sources) ²⁴⁶. Claude Opus 4.8 enables reliable autonomous systems that run unattended for hours, using trigger-agent-verification pattern (corroborated by 2 sources) ³⁵⁸. Claude Code with Higgsfield MCP can produce a fully animated, scroll-driven website with cinematic effects for a fraction of traditional cost (corroborated by 2 sources) ³⁰⁸. Fable 5 is designed for complex, multi-step tasks that take humans hours or days (corroborated by 2 sources) ⁹⁵. Fable 5 can be given a goal, readiness criteria, and limits to autonomously plan, write, test, fix, and deliver a result (corroborated by 2 sources) ⁵³⁸. Sonnet 5 offers Fable-style agent orchestration at a cheaper tier but may not replace Opus based on testing (corroborated by 2 sources) ⁴³⁷. Claude Opus 4.8 can generate fully interactive 3D websites from a simple prompt (corroborated by 2 sources) ⁴⁶⁶. Different frontier models produce distinct creative outputs when given the same product brief in an agentic pipeline (corroborated by 2 sources) ²⁰⁶. By adjusting effort, prompts, caching, batching, budgets, and routing, Fable 5 costs can be halved (corroborated by 3 sources) ⁹⁷. Fable 5 is free until July 7; use it with Opus 4.8 planning to clone software, audit Claude Code setup, review codebases, or build long-horizon apps (corroborated by 2 sources) ⁸³. With careful configuration, Qwen3.6-27B at Q8 quantization can achieve near 100K context on 32GB VRAM (corroborated by 2 sources) ³⁵⁷.[21]

Tooling landscape

Frontier models: Fable 5 (most cited), Opus 4.8, Sonnet 5, GPT-5.5 (Codex), Gemini 3.5/3.6/4 Flash, Gemini Omni, Seedance 2.0, Kling 3.0, Nano Banana Pro, Meshy, Mediapipe, Higgsfield MCP, ClaudeKit's EngineerKit, fable-advisor plugin, GLM 5.2, Qwen3-VL, Gemma4, LongCat, Hermes Agent, Kyutai's Pocket TTS, SynthID. Infrastructure: Two used RTX 3090s (48GB VRAM, ~$1,400) ²⁵¹, Mac Mini with Ollama ^250,169, Ugreen DXP6800 Pro NAS ²¹², reMarkable Paper Pro ²⁴⁶. APIs: Gemini 3.5 Flash free via REST ⁸⁴, GitHub repo of permanently free AI APIs ⁸¹, ClickUp 50 free credits ⁸⁹. Workflow tools: n8n ⁹², ComfyUI ⁹¹, Obsidian ^163,170, Torlink ²⁴⁵, llama.cpp and llama-swap ⁵⁰⁵.[14]

Contested & unverified

Most claims are emerging with single-source corroboration. Notable unverified claims: Fable 5 generating a complete game with 10 levels and web3 integration from a single prompt ³¹¹; using Semantic Similarity Rating with GPT-4o to replicate consumer surveys with 90% reliability ³¹⁰; Fable 5 generating complex 3D animations without external assets ²⁸⁷; AI repurposing workout footage into recurring revenue ²⁸⁶; GPT-5.5 and Opus 4.8 generating HTML canvas animations from abstract concepts ²⁴⁷; Gemini models generating a voxelized human heart ²¹⁴; workflow earning $60-$100 per asset ²¹⁰.[7]

What changed in this versionvs v12

Added new corroborated practices: Fable 5 autonomous planning (538), Sonnet 5 orchestration (437), Opus 4.8 3D websites (466), model output diversity (206), cost halving strategies (97), free Fable 5 until July 7 (83), Qwen3.6-27B context on 32GB (357). Added new tooling: GLM 5.2, Qwen3-VL, Gemma4, LongCat, Hermes Agent, Pocket TTS, SynthID, llama.cpp/llama-swap. Added new unverified claims: Fable 5 game with web3 (311), semantic similarity surveys (310), 3D animations without assets (287), work

+Notable unverified claims: Fable 5 generating a complete game with 10 levels and web3 integration from a single prompt (311); using Semantic Similarity Rating with GPT-4o to replicate consumer surveys with 90% reliability (310); Fable 5 generating complex 3D animations without external assets (287); AI repurposing workout footage into recurring revenue (286); GPT-5.5 and Opus 4.8 generating HTML canvas animations from abstract concepts (247); Gemini models generating a voxelized human heart (214); workflow earning $60-$100 per asset (210).

−Notable unverified claims: Fable 5 generating a complete game with 10 levels and web3 integration from a single prompt (311); using Semantic Similarity Rating with GPT-4o to replicate consumer surveys with 90% reliability (310); Fable 5 generating complex 3D animations without external assets (287); AI repurposing workout footage into recurring revenue (286); GPT-5.5 and Opus 4.8 generating HTML canvas animations from abstract concepts (247); Gemini models generating a voxelized human heart (214); workflow earning $60-$100 per asset (210); structured JSON prompt in Seedance 2.0 (209); Gemini Omni for high-quality AI videos (208); Fable outperforming expectations (207); different models producing distinct outputs from same brief (206); Fable generating app with Seedance transitions (205); local Mac Mini reducing Claude API usage by 70-80% (169); 30-day fitness challenge generating $7,400 per round (168); Claude automating multi-step trading analysis in under 2 minutes (167); Claude Sonnet 5 tokenizer cost changes (165); Anthropic using Obsidian knowledge graph (163); OpenAI employees using Codex agents (158); Claude-built trading bot on Polymarket (102); Claude Code automating legal setup (101); Fable 5 cloning complex app in under an hour (100); switching from Opus 4.8 to Sonnet 5 reducing cost (99); Fable 5 costs halved via adjustments (97); local AI server replacing subscriptions (96); Fable 5 for complex multi-step tasks (95); Sonnet 5 underperforming Opus 4.8 on canvas animation (94); Interactions API chaining models (93); n8n pipeline for faceless YouTube (92); ComfyUI workflow for consistent character (91); pairing planner and executor (88); vibe coding workflow (87); Fable 5 free until July 7 (83); Codex with GPT-5.5 Pro for overnight PRs (82); adding index files to reduce task time (79); structured design workflow with Claude Code (78); zero-headcount company (77); AI-assisted porting of Kubernetes to TypeScript (76); Fable 5 on 'High' setting reducing cost (255); Fable 5's 'Taste' skill (248); Fable 5 as orchestrator with fable-advisor (213); Fable 5 team prompting strategies (211); Fable 5 with Higgsfield MCP for marketing campaign (249); Fable 5 reverse-engineering router firmware (471); Fable 5 building multi-tool command center (473); Fable 5 generating game clones (439); Fable 5 orchestrating hardware tasks (436); Fable 5 finding code improvements Opus missed (435); Claude Code generating complete game in weekend (434); GPT-4o generating fashion portraits (433); 12 open-source projects (432); Claude Opus 4.8 autonomously processing client meetings (359); cluster of small classifiers replacing cloud APIs (356); building MCP servers generating $500-$3K/month (355); orchestrating multiple frontier models via controller (353); prefill speed critical for agentic workloads (352); Sonnet 5 as fast implementer in multi-agent stack (351); Hermes agent autonomously running business (350); Fable model too restricted (349); 3D Gaussian Splatting for real estate (348); Fable 5 meditation game (313); 4 AI agents managing Etsy stores (312); RelayApp using Claude Opus 4.8 (651); GLM 5.2 replacing Claude in Claude Code (650); Claude Fable 5 building custom skill (649); Claude producing browser racing game (648); Gemma 4 12B producing 3D WebGL app (647); three-step AI consulting offer ladder (646); Hermes Agent's web_extract tool (645); accessing Claude Opus 4.8 via Notion trial (644); Claude prototyping and selling AI product (643); Fable 5 porting legacy x86 games to ARM64 (642); Qwen3-VL and Gemma4 for security analysis (610); Fable 5 writing Polymarket arbitrage bot (609); 20 AI agents managed by GPT-5.5 (608); SynthID watermarking (607); Kyutai's Pocket TTS (606); Claude Fable 5 generating hero images (602); Claude Fable 5 automating TradingView (601); Cursor with GPT-5.5 building macOS app (600); prompt library for AI-generated UGC ads (599); Coinbase switching to Chinese models (598); Anthropic finding conscious-unconscious divide (596); Claude generating short-form video scripts (565); Claude Sonnet 4.6 generating viral kids videos (564); Claude Fable 5 best quality animation but higher cost (539); LongCat model for physics simulations (537); Anthropic free workshops (536); Gemini 3.5 Pro generating Three.js solar system (509); making Claude sound like you (508); CLAUDE.md file with project-specific rules (507); GLM 5.2 Max highest score in coding benchmark (506); high-end consumer hardware running 30B-70B models (505); Claude building mini-scraper leading to $11,000 domain sale (504); Fable 5 as planner/conductor (503); OpenScience multi-model research agent (502); Shepherd enabling agents to revert (501); structuring knowledge vault for 100x speedup (475); building character sheet for Seedance 2.0 (474); 'You are Jarvis' prompt (473); Stage 3 AI usage writing loops (472); Claude generating full desktop concept (470); Fable 5 and Opus 4.8 different results (469); GPT-5.6 Sol simpler code than Opus 4.8 (468); seven AI agents automating YouTube channel (467); Claude Opus 4.8 generating interactive 3D websites (466); Claude Fable 5 improving quant trading bot (465); Gemini Omni Flash adapting video clips (464).