Hi all, we’re the team behind TrueOpen. We’ve built an open network for verifiable and private AI inference: independent GPUs serve open-source models, and the network verifies that the agreed model actually ran, with encrypted inputs and outputs. After more than three years of R&D, our public testnet is live.
Try it now: you can run verifiable inference right from the chat on our homepage, https://trueopen.ai
If you run GPUs or maintain open models: node operation and model registration are rolling out on the testnet now. Reply here and we’ll get you in the first wave.
The core of the project is the verification, and we’d value this community’s scrutiny:
- Logprob Majority-Consensus Verification (LMCV): independent verifiers recompute output-token logprobs, ranks, and top-k sets via teacher-forced prefill. In Qwen3-8B and 32B tests, verification cost was about 1 to 8% of generation GPU time.
- Sampled Layerwise Proofs (SLP): commit to all computation chunks, then prove the mandatory ones plus a random sample to cut proving cost.
We have no token and are not doing a coin sale. Settlement uses USDC.
- Code, data, whitepaper, and research papers are all on our GitHub ( TrueOpen.ai · GitHub ) and linked from our homepage.
Feedback on the verification approach is very welcome.