←
inferenceFI23 posts
inferenceFI avatar

inferenceFI

@inferenceFI

The real-time market for verified inference produced on @Tensorcash, built on its rails.

◷ Joined October 2026▧ 4 Photos and videos
inferenceFI avatar
inferenceFI@inferenceFI·Oct 22

This account is now @inferenceFI, formerly @AskTillyAI. inferenceFI will become the commercialization layer for the verified inference produced on @Tensorcash, and we’re looking forward to finding a home for the verified compute running on the network. inferenceFI is being built on the TensorCash rails, using its verification protocol and settlement layer. AskTilly, the consumer app, comes online later at @AskTillyAI.

inferenceFI avatar
inferenceFI@inferenceFI·Nov 17

A market for inference needs two ingredients. Proof, so buyers can trust sellers they have never heard of. And an open market, where spare capacity meets the buyers who need it, at open prices. Welcome to inferenceFI.

inferenceFI avatar
inferenceFI@inferenceFI·Nov 9

With public bids and trades, prices set by real supply and demand, and proof of the model behind every eligible answer, inferenceFI is building the real-time market for verified inference produced on @Tensorcash, built on its rails. Stay tuned.

inferenceFI avatar
inferenceFI@inferenceFI·Nov 9

This lets buyers trade inference just like any other asset: with market orders that run now, limit orders that wait for their price, and in time the possibility to hedge inference costs. Opening the door to what was not possible before.

inferenceFI avatar
inferenceFI@inferenceFI·Nov 9

With this proof, inference no longer has to be priced through GPUs, which never worked well: GPUs get cheaper as they age, and performance depends on many factors. Buyers pay for what they get, and tokens will be priced for what they are worth, by real supply and demand.

inferenceFI avatar
inferenceFI@inferenceFI·Nov 9

The supply comes from @Tensorcash, where mining means running AI models. Miners produce inference and compete for the block reward: two outcomes from one unit of electricity. The protocol brings a crucial factor: mining generates proof of the model behind every eligible answer.

inferenceFI avatar
inferenceFI@inferenceFI·Nov 9

Last month we explained why inference will trade as a commodity. Below, we explain how we are building the market where it does: buyers and sellers trading inference at open prices, with proof of what they get.

inferenceFI avatar
inferenceFI@inferenceFI·Oct 29

As companies use AI for more and more of their work, inference is becoming a fixed cost in every budget. A cost that size needs certainty about what you get, what you pay, and what you’ll pay next month. Imagine a market where every answer comes with proof of the model that wrote it, prices are open, and costs can be hedged like energy. That’s the market inferenceFI is building on @Tensorcash.

TensorCash avatar
⟳ inferenceFI retweeted
TensorCash@Tensorcash·Oct 26

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.

Post attachment
TensorCash avatar
⟳ inferenceFI retweeted
TensorCash@Tensorcash·Oct 19

TensorCash is a proof-of-work chain where the work is AI inference. Miners can sell this inference and get a chance at the next block with the same work. It sets the standard for verifiable inference by attaching a proof of the model used to every answer.

inferenceFI avatar
inferenceFI@inferenceFI·Oct 17

Every model turn came with a proof. A turn is one round: the agent asks, the model answers, the agent acts and reports back. 42 rounds in this session. Each proof covers 256 tokens of what the model wrote, so anyone can check which model actually did the work.

Post attachment
inferenceFI avatar
inferenceFI@inferenceFI·Oct 17

The model ran on a single TensorCash miner, a GPU that mines the chain and answers API requests in the same process. So every answer it gave was also proof of work. 1,207,526 input and 37,063 output tokens came to $0.16 with caching.

inferenceFI avatar
inferenceFI@inferenceFI·Oct 17

The setup: OpenAI Codex CLI connected to the AskTilly API, running Qwen3.8-27B, the model registered on TensorCash mainnet this week. 250k-token context window, one prompt, 42 actions taken on its own, 28 minutes of work.

0:18