Changelog

The evolution of the PRISM engine.

Every release is measured before it ships. Below is the trail — capabilities, performance, and the fixes in between.

v2.7.0July 2026Major Release
  • Direct indexing validated to a 500,000-account book on a single GPU — ~2 min cold, ~25–30× a tuned baseline, matched quality.
  • Real-time dispatch scaled to a 1,000,000-device fleet — ~15 s end-to-end, ~25× faster, comfortably inside the 5-minute market window.
  • Benchmarks reconciled to one consistent evidence set across every page — real-data ceiling now at 75,257 assets.
  • Site-wide refresh with real photography across every solution and application page.
v2.6.0-betaJanuary 2025Early Access
  • Hybrid solver backends — seamless switching between GPU and research QPU kernels.
  • Quantum-addressable memory interface (research track) for hybrid refinement experiments.
v2.5.0Early January 2025Performance
  • Zero-copy portfolio loader — 75,000 assets in <50 ms.
  • Gradient-free optimization kernels for non-convex horizons.
v2.4.0December 2024Major Release
  • Collateral optimization engine.
  • Cloud API security hardening (v3).
  • Tax-alpha calculator for direct indexing.
v2.3.1November 2024Patch
  • Fixed a cardinality-constraint edge case in N=2,000 portfolios.
  • Improved CUDA memory management for dense matrices.
v2.3.0October 2024Feature
  • Initial direct-indexing solver release.
  • Added turnover-limit constraints.

See the latest, measured on your data.

Every capability above traces to a recorded benchmark. Run the current engine against your own universe.

Request a pilot →