Raycaster
Raycaster Eval is a public benchmark platform that publishes complete, scored runs of AI agents across professional domains like legal, finance, and biopharma. Each published run includes the task prompt, input files, full agent execution trace, output artifacts, rubric-based grading results, and cost/token usage data, enabling cross-benchmark comparisons such as M&A legal due diligence with model pass rates and average cost per run.
- Artificial Intelligence
- Developer Tools
- AI Agents
- Data & Analytics
- Software Only
Funding
Raised to date
$500KRaised to date based on public sources. This may differ from the amount the company actually raised and is based only on what is publicly available on the internet.
Founders
Founder details are not available yet.
Product
Problem
AI agent performance is often described with vague claims or proprietary benchmarks that cannot be independently verified. Teams building agents for professional workflows lack access to concrete, reproducible evaluation data showing how models actually perform on real-world tasks, what they cost per run, and where they fail against expert rubrics.
Solution
Raycaster Eval is a public benchmark platform that publishes complete, scored runs of AI agents across professional domains. Each published run includes the task prompt, all input files, the full agent execution trace, output artifacts, and rubric-based grading results, alongside cost and token usage data. The platform organizes runs by domain and benchmark, offering cross-benchmark comparisons such as M&A legal due diligence where users can see pass rates, average scores, and average cost per run for models like GPT, Gemini, and Kimi. Raycaster Eval also provides downloadable task pages and trace inspections, making agent performance concrete, inspectable, and reproducible for anyone building or procuring agent systems.
Target Audience
AI agent developers, evaluation teams, and enterprises that need to benchmark model performance on professional workflows or make purchasing decisions based on reproducible evidence. The platform also serves academic and industry benchmark maintainers looking for public hosting and comparison infrastructure.
Features
- Published runs with full traces including prompts, files, execution logs, and rubric grades
- Cross-benchmark category pages comparing model pass rates, average scores, and cost per run
- Domain hubs spanning biopharma, energy, healthcare, legal, finance, documents, and model development
- Multiple third-party benchmarks including APEX-Agents (Mercor), SpreadsheetBench 2 (Renmin Univ. & Aptura), and OfficeQA Pro (Databricks)
- Company Worlds providing structured environments like Havenor Therapeutics and Broad Institute scenarios
- Rubric checks shown per run with specific pass/fail items, such as "States Falcon Prevent's Max Refund Amount is $0.00"
- Live public product surface for inspecting full runs through an interactive chat/trace viewer