Priors adds RWE-Bench engineering leaderboard

While LLMs routinely conquer software coding benchmarks, physical engineering remains an unsolved frontier where top models struggle with real-world constraints. Adding RWE-Bench to customizable, weighted indexes gives engineering teams a realistic assessment of physical simulation capabilities paired with practical cost economics.

• –Engineering capability gap: Leading model Fable 5.2 max scores 62% while GPT-6 Astra reaches 37% and Opus 5.5 hits 29%, showing that CAD, thermal simulation, and physical design remain major hurdles for frontier models.
• –Subscription versus API cost disparity: Priors calculates that running Fable 5.2 max under a Claude Ultra subscription costs $4.22 per task compared to $115 per task at standard API pricing.
• –Domain-specific index weighting: Hardware and robotics developers can blend RWE-Bench with coding and reasoning metrics to create personalized leaderboards tailored to industrial engineering workflows.
• –Programmatic eval access: Priors exposes the benchmark data through its CLI, HTTP API, and MCP server, allowing automated agent pipelines to dynamically route physics-heavy tasks to optimal models.