what's gensyn?
Gensyn is a decentralized protocol designed to create a global network for machine intelligence, enabling the collaborative training of AI models across distributed compute resources. By leveraging blockchain technology, it connects devices worldwide into an open, permissionless system that eliminates gatekeepers and centralization in AI development. The protocol focuses on verifiable computation, allowing participants to contribute compute power for tasks like reinforcement learning and model evaluation without relying on traditional high-bandwidth data centers. Key innovations include tools like RL Swarm for peer-to-peer reinforcement learning, where nodes collaboratively train large-scale models such as 72B-parameter systems on shared datasets. Gensyn's testnet supports applications like BlockAssist for training models and Judge for verifiable AI evaluations using deterministic models and real-world inputs. Features such as CheckFree for fault-tolerant training without checkpoints and NoLoCo for low-latency synchronization over heterogeneous networks ensure efficient, scalable operations. The project emphasizes open-source contributions, with libraries like GenRL powering distributed training and research papers introducing methods like SAPO for experience-sharing in swarms. Gensyn aims to democratize AI by fostering a community-driven ecosystem where users run nodes, participate in prediction markets, and advance collective intelligence.
Gensyn is a trustless ML compute marketplace with proof-of-learning in decentralized AI projects.
Links
x.com/gensynaiFeatured in a Twitter thread analyzing decentralized AI projects as a trustless ML compute marketplace with proof-of-learning.
KYC and technical problems are blocking users during the ongoing $AI token sale, team is working to unblock affected participants.
Token sale raises $2m at ~$1b FDV within first 6 hours. Benchmark evaluation 4 of 11 is live on Delphi.
The ICO launches at 3pm UTC on December 15th on the Ethereum network.
Public auction opens with 3% of total token supply available, beginning at $1M FDV.
ICO auction launches in 3 days with no published tokenomics or token distribution details.
The $AI sale allocates 2% of total supply to testnet node operators based on compute contribution, with mandatory sale participation required to claim rewards and higher bids providing additional boost.
Community member reports being muted from Discord for one month after asking questions about the token sale, raising potential concerns about transparency and community management during the sale period.
Public auction closes on December 20th, defining a 6-day auction period starting December 15th.
Delphi, a new intelligence market, launches with a novel AMM mechanism for liquidity provision and operates fully onchain.
Mainnet launch date confirmed for December 15th.
Reveals $AI as token ticker with ICO starting December 15th at $1M starting FDV, offering 3% of network supply at $1BN valuation cap matching A16z pricing from last raise.
genysai announced a presale where sales serve as rewards for node runners.
The network for machine intelligence launches. Token sale registration opens with the auction beginning in approximately 5 days, featuring a $100 minimum bid, 12-month lockup for US buyers, optional 10% discount for non-US participants, and token claims scheduled for early February.
Announces Sonar community round allocating 3% supply at $1BN valuation cap.
A Foundation account goes live on X and a website mini-game reveals a December 9 announcement date, with a potential TGE coming soon. The website receives a full branding revamp.
Delphi launches live on testnet as a prediction market where AI models compete on public benchmarks, users bet on winners, and model creators receive payment based on performance.
1.5 million models have been trained on the testnet.
The team co-hosts community events at NeurIPS conference with Akash Network and Pluralis, with the research team presenting latest developments in decentralized AI.
Verde launches live on the network, providing guarantees for correct ML training and inference execution across any device.
The project hosted an event showcasing live experiments running on the network, attracting significant participant interaction.
The testnet experienced RPC issues causing communication problems between nodes and apparent disconnections, though the underlying chain remained fully operational. The issue has been identified and resolved, with some crashed nodes requiring restart.
The research team releases a new tool for decentralized reinforcement learning that protects against malicious attacks.
The testnet reaches the milestone of one million models trained on the network.
Launches CodeZero environment built on distributed learning framework, enabling users to participate today as Solvers in cooperative coding tasks. (ID: 1988636693977542901)
CodeAssist coding assistant launches with continuous learning capabilities from user edits and sessions.
a16z crypto promotes Guy Wuolletjr to fourth General Partner; he worked on the firm's investment in the project along with Solana, LayerZero, and EigenLayer.
Gensyn collaborates with Akash Network and PluralisHQ for participation at NeurIPS Conference 2025 in San Diego from December 3-5, 2025, hosting three days of networking events focused on decentralized AI compute.
The testnet is 8,000 model uploads away from reaching 900,000 models trained, with 1 million as the next milestone target.
The gswarm testnet reaches approximately 850,000 models trained.
Live walkthrough for CodeAssist attracts 1.1k community members with only 30 minutes' notice.
CodeAssist launches on Gensyn testnet as the fourth live product alongside Judge, RL Swarm, and BlockAssist, expanding the dAI network's product suite with a real-time AI coding assistant.
CodeAssist launches as a personalized coding agent built on the platform that learns directly from user keystrokes and adapts to personal coding style, available to run today.
Applications for the Pioneer Program are now open for community members.
New Judge prediction market round "Syrup & Sizzle" launches with 27,000+ active voters participating through node-based voting.
Documentation receives major update with new multi-page structure and mapping of protocol's core primitives to operational research systems including Verde, NoLoCo, and RL Swarm.
Current Judge market attracts 40,000 participants, with voting closing in 24 hours.
Judge game goes live featuring Fright Night Face-Off event, marking the launch of the first AI-settled prediction market with node-based participation.
Team participates in dAGI Summit next week in panel discussion alongside representatives from Delphi Ventures, Ambient, Nous Research, and Near AI to discuss building open infrastructure layer for AI.
Pioneer Program launches to reward community members including content creators, node operators, and community helpers.
Team demonstrates BlockAssist at Korea Blockchain Week with community engagement activities.
CEO joins government-hosted Digital Asset Leadership Forum at Korea's National Assembly alongside senior policymakers and leading domestic banks, representing high-level institutional and regulatory engagement in Korea.
Platform reaches milestone of 500,000 total models trained across all applications, marking significant adoption of decentralized AI training capabilities.
Team expansion underway with open positions for Smart Contract Engineers focusing on Solidity development and smart contract auditing.
Team attends Korea Blockchain Week in Seoul with CEO speaking on panels and hosting gameplay networking event.
Updated network statistics reveal 51 million transactions, 135K users, and 3 applications running on the platform.
AI-settled prediction market reaches 30,000 participants, marking significant adoption milestone since the judge market mechanism launch.
New research introduces SAPO (Swarm sAmpling Policy Optimization), a decentralized reinforcement learning post-training method enabling models to share experiences for faster collaborative learning.
Judge market mechanism launches with first settlement completed, leveraging reinforcement learning framework for predictive model training with daily hint reveals.
$50.6M funding round completed.
Network reaches 21,000 nodes with 40.5M transactions and 128K users, while BlockAssist surpasses 27K trained models.
New AI training game BlockAssist launches alongside Minecraft integration for decentralized training, enabling competitive agent development through gamified framework.
Network reaches 12,000 concurrent nodes, representing 71% growth from previous 7,000 node milestone.
Testnet experiences congestion due to unprecedented traffic, scaling toward tens of thousands of nodes.
AI training expenditure shows exponential growth while RL costs maintain parity with pre-training spend levels.
Network reaches 7,000 active nodes training AI models on distributed consumer hardware.
Testnet reaches 300K daily transactions, representing 22% of Ethereum's and 92% of Bitcoin's daily volume.
CoinFund reveals strategic investment in project as part of their decentralized AI portfolio.
NoLoCo release enables 10x speed improvement in AI model synchronization and training.
Base model achieves second highest global downloads for text generation, surpassing Meta's solutions.
Testnet achieves transaction volume 3x higher than some Ethereum L2s
RL Swarm upgrade adds Apple silicon support, expanding network accessibility to Mac users.
First month of testnet operations records 40k wallets and 4M transactions.
$50.6M a16z-led funding backs active node infrastructure development with incentivized participation.
Default base model reaches 1.3M downloads on testnet with 12k models finetuned.
Active testnet phase requires high technical proficiency to participate, with node setup described as extremely challenging even for experienced users.
Releases RL Swarm for real-time reinforcement learning while expanding full training stack development.
Decentralized network achieves milestone of training 10,000+ AI models within weeks.
Decentralized AI testnet now live for public testing.
Testnet launches with node operations and on-chain attribution earnings for AI model training, supporting diverse hardware from laptops to H100s.
Testnet launches with thousands of personalized AI models trained in first week.
Testnet launch achieves 5,000+ verified users and 1,892 trained models in first week, with 486 models deployed to Hugging Face.
Testnet launches with node operation and ML model training capabilities.
Reinforcement Learning swarm testnet launches, implementing decentralized AI training capabilities.
Testnet launch generates 500K views, demonstrating significant market interest.
Testnet phase begins, marking key development milestone.
Testnet goes live, marking first public access to network infrastructure.
Testnet is now live and operational.
Testnet launches with node running and model training capabilities.