GENSYN

• Gensyn is a network for distributed machine learning training, letting anyone contribute compute to training runs.

• Its core technical contribution is verifying that submitted training work was actually performed, without repeating it.

• The token is used for payments, staking and governance.

• Total supply is 10 billion, allocated 40.4% to the community treasury, 29.6% to investors, 25% to the team, 3% to the public sale and 2% to testnet rewards.

• Staked as collateral by compute providers against submitting invalid work.

• Investor and team allocations are locked for the first year after the April 2026 launch, then unlock linearly over two years.

Pros:

• Verification of training work is the hard problem in decentralised ML, and Gensyn attacks it directly.

• Backed by a16z in one of the larger rounds in the category.

• Research output is substantive rather than promotional.

Cons:

• Training verification remains computationally expensive relative to the work it verifies.

• Distributed training over the public internet is bandwidth-constrained in ways that matter for large models.

• Long development timeline before a production network.

• Exposure to the most technically ambitious approach in decentralised machine learning.

• Verification research has value independent of the token.

• Weigh the gap between research progress and production usage.

• Probabilistic proof that training work was genuinely performed, which is what makes paying untrusted compute providers possible at all.

• Paying strangers to train a model requires knowing they did the work, and re-running it to check defeats the purpose.

• Probabilistic verification checks a fraction of the computation and makes cheating unprofitable.

• Founded by Ben Fielding and Harry Grieve.

• Backed by a16z crypto, CoinFund and Canonical Crypto.

• Gensyn held a public token sale on Sonar in December 2025, offering 3% of supply by auction. It raised about $16.1 million at roughly $0.0473 per token, a fully diluted valuation of about $473 million.

• Verification overhead may remain too high for economic viability.

• Bandwidth limits distributed training of large models.

• Centralised compute continues to improve in cost and availability.

• Long timeline from research to revenue.

Gensyn's papers frame the obstacle to decentralised machine learning as verification rather than capacity: there is abundant idle compute, but no way to pay an anonymous provider for training work without a means of checking they performed it, and re-running the computation costs as much as doing it. The proposed solution combines probabilistic proof-of-learning, graph-based pinpointing of where two claimed computations diverge, and a staking game in which challengers profit from catching invalid work. The design is explicit that verification is probabilistic rather than absolute: the aim is to make cheating unprofitable in expectation rather than impossible.