Logical Intelligence's AI coding agent Aleph has achieved top scores on four major formal reasoning benchmarks, marking a significant advance in formally verified code generation. The system solved 99.4% of PutnamBench problems, achieved 94% success on VeriSoftBench, and recorded state-of-the-art performance on LeanEval and Verina. Founded by Eve Bodina, the San Francisco startup develops energy-based reasoning models for mission-critical systems. Unlike traditional AI that generates plausible outputs, Aleph produces machine-checkable proofs in formal verification environments, addressing concerns around hallucinated code and unverifiable outputs. The agent is currently used in production verification workflows, including work with the Ethereum Foundation's cryptographic libraries. Logical Intelligence plans to launch a beta version later this year, targeting critical infrastructure, industrial software, semiconductor design and finance sectors.

Logical Intelligence has announced that Kona 1.0, its energy-based reasoning AI model, will enter pilot programmes with partners in energy, advanced manufacturing and semiconductor industries this quarter. The company has released a live demonstration featuring sudoku challenges against large language models, with chess and Go demonstrations planned. The San Francisco-based startup has appointed Yann LeCun, former chief AI scientist at Meta, as founding chair of its Technical Research Board, and Patrick Hillmann, formerly of General Electric and Binance, as Chief Strategy Officer. Unlike language models that predict likely answers, Kona learns by recognising and correcting its own mistakes through energy-based optimisation. CEO Eve Bodnia described it as showing "first credible signs of AGI", whilst emphasising that AGI will require multiple model types working together.