Most AI tools lock you to a specific model from a specific lab. Sourcetable doesn't. We're model-agnostic: always the frontier, always GPT-5 today, always whatever's best tomorrow. Here's what that means for your work.
Andrew Grosser
June 3, 2026 • 6 min read min read
There's a subtle trap in how most AI tools are sold. You pay for access to a model — GPT-5, Claude, Gemini. That model is the product. Which means when a better model drops, the product you paid for is no longer the best available. You either pay more, switch subscriptions, or accept that your analysis is running on yesterday's intelligence. Sourcetable is built differently.
Sourcetable's core insight is simple: for analytical work, the AI model should be infrastructure — like the database underneath your application. You wouldn't lock your production database to a specific version and refuse to upgrade when a better one drops. You'd upgrade, because better infrastructure means better performance.
Sourcetable treats AI models the same way. We're model-agnostic: we evaluate the frontier continuously and run whichever model is currently leading. Today, that's GPT-5. When something better arrives — from OpenAI, Anthropic, Google, or anyone else — Sourcetable upgrades automatically. The model is the infrastructure. The product is what you do with it: analyze your data, build financial models, generate reports, and make better decisions.
Sourcetable scored 100% on the Vals.ai finance agent benchmark — the same benchmark where Claude Opus 4.5 scored 67% and most general-purpose AI tools don't participate at all. That 100% score reflects what frontier-model AI looks like when it's purpose-built for analytical work: correct DCF models, accurate Monte Carlo simulations, reliable natural language queries against real databases.
But here's the compounding advantage: as the frontier advances, Sourcetable's scores improve automatically. Better models mean better formula generation, sharper analysis, more accurate financial projections — all running on the same data connections you set up once.
Sourcetable works with your real data. Not uploaded CSVs that expire at the end of a chat session — live connections to the data that runs your business. Postgres, MySQL, Salesforce, Stripe, Shopify, Google Sheets, Airtable, Notion, HubSpot — 100+ connectors, all accessible through natural language queries.
Build a revenue model once and it pulls fresh Salesforce data every morning. Create a dashboard once and it queries your Postgres production database in real time. Write a formula once with AI assistance and it lives in your spreadsheet permanently. This is what persistent AI work looks like — and it gets smarter automatically every time the frontier advances.
The practical difference between frontier and second-tier models shows up in the work. A financial model built on the frontier is more likely to get the edge cases right. A natural language query run on frontier intelligence is more likely to return exactly what you asked for. A dashboard built with frontier AI formulas is less likely to have errors you find three months later.
This isn't abstract. It's the 33-point benchmark gap made concrete in your daily work: a slightly sharper analysis, a formula that works correctly the first time, a report that doesn't need a second round of review. Over a year of daily analytical work, those marginal improvements compound into significantly better outcomes.
Sourcetable's frontier model commitment: