Post #25581

$POD / Dolphin AI 4M+ monthly downloads on Hugging Face. The Dolphin LLM family (built by Eric Hartford / Cognitive Computations) is one of the most-used open-source model families in the world: and it now has its own token + decentralized inference network Who's backing & talkinga bout it — a16z funded the original Dolphin training compute — Marc Andreessen has personally endorsed Dolphin multiple times. May 2026: NYT-style piece covered Marc using Dolphin as his personal chatbot — Erik Voorhees / Venice (VVV), Dolphin built the model that IS Venice's default product — Bindu Reddy (Abacus AI) publicly called Hartford "one of the most prolific open-source AI contributors" — vitalik spoke about it, uses it why's it so big — Dolphin-Mistral-24B-Venice-Edition = the default model on Venice.ai ("Venice Uncensored") — Venice does 1M+ daily API inference requests — Venice (VVV) market cap is 10–15× larger than POD's — POD is the upstream model provider for VVV's flagship product, at a fraction of the valuation new stuff — Venice Uncensored 1.2: vision support, 4× larger context, stronger tool-use — Trained on Targon (Bittensor Subnet 4), 1,500+ H200s, $70M+ NVIDIA hardware, 20B+ paid inference tokens/day. $TAO exposure baked in — OpenRouter integration: free tier = viral developer distribution — 2.2% censorship refusal rate vs 71% Claude / 64% GPT-4o-mini. Most uncensored production model in existence. Peer-to-pool decentralized inference. Gamers/GPU owners stake POD, serve inference, earn POD. Unlike Akash or io.net (rent-by-session), nodes can join/leave freely. — V1 live since October — V2 launching in 4–6 weeks: enterprise GPU batching + public API — Roadmap: image, video, audio, Whisper transcription, distributed RL training, synthetic data engine, *ame-dev SDK for in-game AI NPCs — Even CPU nodes can earn POD via web scraping work in development, Dolphin X1 235B (Qwen3 base) and Dolphin X1 405B (Llama-3.1-Tulu-3 base), competitive with frontier-class open models. tokenomics of $POD — Supply: 500M — Mcap: $35.9M | FDV: $56.8M — Top-2 wallets = locked contracts (88.8%), supply is genuinely thin — Real liquid float ≈ 50M tokens — only on Uniswap, Base & LBank market context: AI inference TAM going $106B (2025) → $255B (2030) Compare POD ($36M) to: Bittensor (multi-B), io.net (hundreds of M), Akash (hundreds of M), Render (multi-B), Aethir (hundreds of M). 1–2 orders of magnitude smaller than every comp. tldr: — Real founder (Hartford) — Real product (4M+ monthly downloads = real users right now) — Real upstream relationship to a $500M+ token (VVV) — Real Bittensor exposure (Targon) — Real Marc Andreessen mindshare — Real network shipping (V2 in weeks) At $36M mcap with this stack, the asymmetry is obvious