Performances
RD Popularity Index
RD Growth Index
| 1h | 24h | 7d | 30d | 60d | 90d |
|---|---|---|---|---|---|
| 0.61% | 7.43% | 4.16% | 120.18% | 116.86% | -- |
Comparison
Top Holders
| Address | Token | Quantity | Percentage | Value |
|---|---|---|---|---|
| 0xec29...f988 |   POD | 256,876,764.08 | 51.38% | $160.41M |
| 0x8e35...bef6 | POD | 160,106,441.31 | 32.02% | $99.98M |
| 0xc19a...7c0a | POD | 38,265,746.93 | 7.65% | $23.89M |
Follow Updates
Follow Lists
People
About Dolphin
Dolphin is a decentralized inference network that aggregates idle consumer GPUs into a shared computing pool. Its core business allows GPU owners to run lightweight node software on Windows or Linux, contributing computing power while earning POD tokens, while developers and users can access AI model inference services at lower cost.
Its main competitive advantage lies in the Peer-to-Pool architecture, which differs from traditional 1:1 matching models by allowing nodes to join or leave freely without long-term commitment. The network employs software integrity checks, validator nodes, and cryptoeconomic bonds to prevent malicious nodes from tampering with models or returning incorrect results, ensuring output quality. Additionally, by leveraging Nvidia's market segmentation—consumer GPUs are significantly cheaper than data center GPUs but deliver comparable inference performance—Dolphin reduces costs by approximately 30%.
The market pain point addressed is the high cost and centralization of AI inference. Traditional cloud services are expensive and raise privacy concerns, while existing decentralized networks often require strict node uptime and lack effective verification mechanisms. Dolphin's flexible participation model and multi-layered verification system offer a more accessible and trustworthy alternative.
In the past six months (March to September 2026), Dolphin launched V1 of its network in October 2025 and has been preparing V2, which will introduce batching for enterprise-grade throughput and a public API. The roadmap also includes sharded inference for large models (e.g., splitting a 235B model across multiple GPUs), distributed LoRA training, and expansion into audio generation. However, the network still faces challenges common to distributed systems, such as variable node performance, complex result verification, and regulatory uncertainty around open AI models.
Dolphin Network has progressively built a decentralized inference network since its $886K seed round on June 20, 2024. Key milestones include: creating the Dolphin Mistral 24B Venice Edition model serving over a million Venice.ai users; surpassing 5 million monthly downloads on Hugging Face; launching Dolphin Inference Network v1 in October 2024 to utilize idle consumer GPUs.
The roadmap outlines long-term nodes: V2 introduces batching for enterprise throughput; expanding image, video, audio and transcription model support; implementing distributed LoRA training (validated via Llama3.1 405B); advancing distributed reinforcement learning and full-parameter SFT; culminating in large-scale distributed pre-training, complemented by model creation suites and tokenomic incentives.
Positive:
Protocol revenue 100% used to buy back POD, directly offsetting inflation; POD surged 14x in a month, market cap ~$15M; staking xPOD earns auto-compounding, AI inference credits, and premium subscription benefits; node staking aligns interests, borrowing DeFi mechanisms like Curve; team 12-year vest, no equity, 99% revenue to protocol.
Negative:
Public API not yet open, external revenue unrealized; supply suppressed at 800 GPUs with low rewards; buyback split ratio undecided; API may slip past September; independent benchmarks may fall short of claimed 88%.
Eric Hartford (Core Contributor)
- Founder of Quixi AI, creator of the Dolphin and Samantha open-source AI model series.
- Previously Chief Scientist at Lazarus AI; Senior Applied AI Researcher at TensorWave; Senior Applied AI/ML Researcher at Abacus.AI and Convai.
- Earlier career: Senior SWE at Microsoft, Staff at eBay, SDE at Amazon.
- Over 30 fine-tuned model releases on Hugging Face (e.g., Dolphin Llama 3, Mistral, Qwen2).
- Active on Twitter (@QuixiAI) and LinkedIn (linkedin.com/in/ehartford).
Note: No other co-founders or C-level executives were identified from the provided sources.