Open models erode Harmonic pricing power

Open source turns mathematical AI from a premium product into a capability that serious teams can try for free and tune themselves. That matters for Harmonic because its enterprise pitch depends on charging for verified reasoning, while DeepSeek-Prover-V2 gives researchers and internal platform teams a Lean 4 theorem prover they can download, inspect, and build around. Harmonic still sells a cleaner product layer, but open source lowers the floor on what customers must pay for.

• DeepSeek-Prover-V2 is not a toy release. It ships in 7B and 671B versions, is built for Lean 4 formal theorem proving, and reports 88.9% pass rate on MiniF2F. That gives engineers a real baseline they can self host instead of starting with a paid vendor.
  1 github
• Harmonic packages more than a raw model. Aristotle takes natural language math problems through app, web, and planned API workflows, translates them into Lean 4, checks each proof step, and returns both a plain language explanation and proof code. That product wrapper is where pricing power has to migrate as base models spread.
  2 sacra
• The specialized field is crowding quickly. Axiom Math is also building an AI mathematician around detailed proof generation and reached a $1.5B estimated valuation in February 2026. Once open models win developer adoption and startups raise around the same workflow, benchmark leadership alone becomes less defensible.
  3 sacra 4 sacra

The market is heading toward a split where open models supply the reasoning engine and companies like Harmonic compete on verification infrastructure, workflow, and trusted deployment in finance, engineering, and safety critical software. The winners will be the teams that make formal math useful inside real production systems, not just the teams with the best standalone benchmark score.