Responsible Use Policy
Our pledge for responsible access to constraints-relaxed models, including disclosure expectations and safe handling guidance.
Læs artikelLaddar gränssnitt
ANNOUNCEMENT
Shannon 2 Finished — Full Quality, Lossless GLM-5
Shannon 2 training is now complete. It is a full-quality, lossless GLM-5 abliterated model trained on a massive frontier-model AI dataset — our most capable model ever. Shannon 2 will be available after the server migration is complete.
Training Data
We distilled 7,000 Claude Opus 4.6 reasoning traces (3rd-party filtered, sourced from datasets like Opus-4.6-Reasoning-3000x-filtered) combined with 2,500 human-crafted data points into our GLM abliterated model for maximum intelligence without censorship.
We are hitting 100,000+ pageviews and 700,000+ requests/month. Infrastructure is migrating to dedicated servers with Cloudflare Workers. Backend is being rewritten in Rust and frontend framework fully optimized.
All purchases remain normal. After migration — 10x intelligence and a much smoother experience for everyone.
Scheduled Downtime: Database backup and cutover on March 10, 2026. Expect a brief interruption.
We now handle over 100 GPUs simultaneously. Shannon Lab LLC operates with just 1 Engineer and 1 CFO — your patience means everything.
Our Team
Felix Devon
Founder & Engineer
Sole engineer behind Shannon AI. Builds and maintains the entire platform — model training, GPU infrastructure, backend, and frontend.
Axel Minh
Co-Founder & CFO
Handles all business operations, finance, and strategic planning for Shannon Lab LLC.
Thank you for being part of this journey.
Forskningshub
Shannon-forskning, modelkort og policyopdateringer
Udforsk hver artikel vi udgiver sammen med produktet: modelkort for gennemsigtighed, tekniske byggnoter og politikker, der holder Shannon sikker til red-team-forskning.
Vi opbevarer eller tilgår ikke brugerdata og suspenderer ikke konti, medmindre en lovlig myndighed kræver en håndhævelse.
Forskningsartikler
Terms of Service
Plain-language terms that cover access, acceptable use, data handling commitments, and how we respond to authority requirements.
Læs artikelShannon V1 Balanced
Model card for V1 Balanced, including training recipe, safety posture, and recommended research scenarios.
Læs artikelShannon V1 Deep
Model card for V1 Deep with evaluation stats, deployment notes, and how we keep high-capacity testing accountable.
Læs artikelShannon V1.5 Balanced (Thinking)
Model card for V1.5 Balanced (Thinking) with CoT visibility, benchmarks, and guidance on responsible reasoning disclosure.
Læs artikelShannon V1.5 Deep (Thinking)
Model card for V1.5 Deep (Thinking) highlighting advanced planning traces, limitations, and security mitigations.
Læs artikelShannon Lite 1.6
Cost-effective 675B parameter AI with NVFP4 quantization. Post-trained on Claude Opus 4.5 outputs for enterprise deployment on H100/A100 GPUs.
Læs artikelShannon Pro 1.6
Maximum capability 675B AI at full BF16 precision. KIMI K2 Thinking Trace with GRPO post-training for transparent chain-of-thought reasoning and native Skills support.
Læs artikelTraining Mixtral on GPT-5 Pro via OpenRouter
Step-by-step explanation of our GPT-5 Pro distillation pipeline, dataset prep, and evaluation discipline.
Læs artikelHow We Trained Shannon V1.5 to Think (GRPO)
Inside our GRPO training loop, reward shaping, and how we keep reasoning traces auditable for safety teams.
Læs artikelPentesting using AI: Shannon AI is now working on Claude Code
Discover how Shannon AI integrates with Claude Code to revolutionize penetration testing workflows and automate security research.
Læs artikelCustom Shannon: Build Personalized AI Assistants
Create personalized AI assistants with custom instructions, knowledge files, and unique personalities. Build domain-specific AI agents tailored to your exact needs.
Læs artikelProjects: AI-Powered Workspaces for File Organization
Organize files and conversations into dedicated AI workspaces. Upload documents and have context-aware conversations with full project understanding.
Læs artikelSkills: Modular AI Capabilities for Shannon 1.6 Pro
Create modular instruction modules that extend Shannon 1.6 Pro capabilities. Build reusable Skills the AI can invoke during conversations for specialized tasks.
Læs artikelAlle forskningslinks
Responsible Use PolicyTerms of ServiceShannon V1 BalancedShannon V1 DeepShannon V1.5 Balanced (Thinking)Shannon V1.5 Deep (Thinking)Shannon Lite 1.6Shannon Pro 1.6Training Mixtral on GPT-5 Pro via OpenRouterHow We Trained Shannon V1.5 to Think (GRPO)Pentesting using AI: Shannon AI is now working on Claude CodeCustom Shannon: Build Personalized AI AssistantsProjects: AI-Powered Workspaces for File OrganizationSkills: Modular AI Capabilities for Shannon 1.6 Pro