TensorCash

Network difficulty on TensorCash reached a new all-time high today. More AI inference is running to secure the chain than ever before, and every block still comes with a proof of the model that produced it.

On TensorCash, two people can trade directly, with no one else in between. The funds stay in their own wallets until the trade settles, removing the risk of an exchange losing or freezing them. This video follows one of those trades, from the offer to the block it settled in. One unit of a call option on the network’s difficulty was traded for 0.9 TSC. The option and the TSC moved in one transaction, straight from one wallet to the other. Each trade on the protocol is trustless, and neither side can end up paying without receiving. All in an ordinary transaction, with no escrow, no multisig and no trace of the negotiation, which keeps it cheap and private.

TensorCash v1.2.4 activates today at block 33,378, expected between 09:00 and 11:00 UTC. From that height, nodes run their own validity check on proof-of-inference blocks. Verifiers and bcore block producers who haven’t upgraded yet: the upgrade is mandatory, so do it before then. Pool verifiers: download both pool-verify-pack archives, verify them against SHA256SUMS and extract them into the same directory.

Improved verifier performance for larger models and longer context windows, reviewed together with several miners. TensorCash v1.2.4 activates on Wednesday 28 October at block 33,378, expected around midday UTC. Verifiers and bcore block producers: the upgrade is mandatory, so upgrade before that height.

An independent benchmark by @boblabs_eu showed that mining on TensorCash adds at most 0.3% GPU time to inference. Qwen3.8-27B ran on three GPUs over a full agentic session, once without mining and once with every request mined. GPU time changed by −1.1% to +0.3%. At close to zero overhead cost, a miner can serve inference together with a proof of the model that wrote it, and compete for the block reward at the same time.

Tomorrow at 12:30 ET, early contributors @samjavadi and @enrico_otta join @MCGlive to talk about TensorCash and answer questions about the protocol.

INFERENCEFI has been registered by @inferenceFI as a native asset on TensorCash at block 32,453. Its issuer document, the INFERENCEFI Issuer Charter, is attached to the asset’s Issuer Control Unit (ICU) and describes what the asset is for and the rules it follows, with every version staying recorded on chain. inferenceFI plans to use the native asset to publish index values on chain, and to create child assets under INFERENCEFI.<name>, such as markets, contract legs, outcome tokens and tranches, each with its own terms. Under its rules, inferenceFI publishes what each index measures, and how it is calculated, no later than its first value. The chain confirms that each value came from inferenceFI, and once published, a value can’t be changed. Beyond that, any native asset on TensorCash can, among other things, carry legal documents and holder-only disclosures, require zero-knowledge KYC, let holders vote on changes, and be traded through on-chain spot, repo and forward contracts.

Two models can now be mined on TensorCash: Qwen3-8B and Qwen3.8-27B. The newly added 27B has more than three times the parameters, making it more capable. That also means each 256-token window takes more compute, so a miner would get fewer attempts at a block. To make up for that, and keep both models worth mining from a block reward perspective, its difficulty is set lower. A miner’s chance at a block stays roughly the same on either model.