The Frontier Model Problem: GPT-5 Is Today's Best. What About Tomorrow?
The frontier keeps moving. GPT-4 was the best — until it wasn't. Claude 3 Opus. GPT-4o. Gemini Ultra. Claude Sonnet 4. GPT-5. Every few months, the leaderboard resets and everyone scrambles to upgrade. Sourcetable solves this permanently.
Andrew Grosser
June 3, 2026 • 8 min read min read
Here's a question worth sitting with: when did you last do an analysis and feel completely confident you were using the best AI available? Not 'good enough' AI — the actual frontier. The model at the top of the benchmark leaderboard right now, today, running on your specific task. If you can't answer that with certainty, you have the frontier model problem.
A Brief History of 'The Best'
In early 2023, GPT-4 was definitively the best. Every serious user migrated to it. Then in mid-2023, Claude 2 showed strong results on reasoning tasks and the debate began. In 2024, GPT-4o arrived with faster responses and multimodal capabilities. Then Gemini Ultra launched with Google's knowledge depth. Then Claude 3 Opus outperformed on long-context tasks. Then Claude Sonnet 4 closed the gap everywhere. Then GPT-5 dropped and reset the clock.
Each transition took weeks to fully register. Benchmarks appeared, Twitter argued, product teams updated their model lists, and users scrambled to upgrade their subscriptions. And in every transition window, some portion of serious AI users were running on the model that used to be the best — not the one that is.
The Transition Tax
Every model transition costs you something. If you're using Claude Pro and GPT-5 drops with a significant improvement on financial analysis tasks, your options are: pay for ChatGPT Plus on top of Claude Pro, cancel one and risk losing access when the pendulum swings back, or stay on Claude and accept that your analysis is running on second-best intelligence.
None of those options feel good. And they shouldn't — you shouldn't have to make this choice at all. The model churn is a problem created by a fragmented market where each AI lab is competing for your subscription, not for your best work.
What the Benchmarks Actually Tell You
The Vals.ai finance agent benchmark is one of the clearest measurements of what model quality means for analytical work. These are real financial tasks: DCF modeling, portfolio optimization, natural language queries against databases, regression analysis. Not word games. Not creative writing. The actual work that financial analysts and data professionals do every day.
Sourcetable scored 100%. Claude Opus 4.5 scored 67%. That's not a minor variation — it's a 33-point gap on the work that matters. And that gap was measured at a specific point in time, on specific tasks. As the frontier moves, the gap between whoever is leading and whoever isn't widens and contracts. The only safe position is to always be on the frontier.
Sourcetable's Answer: Always the Frontier, Automatically
Sourcetable is built on a model-agnostic foundation. We don't pick a lab and commit to it — we always use whichever model is leading benchmarks right now. Today, that's GPT-5. When something better arrives, Sourcetable upgrades automatically. No announcement emails. No subscription changes. No thinking required on your end.
This matters even more because Sourcetable's AI runs on your actual data. Connect Postgres, MySQL, Salesforce, Stripe, Shopify, or any of 100+ data sources — and every time the frontier advances, your data analysis gets smarter automatically. A financial model you built six months ago runs on better intelligence today than when you built it. That's what it means to not have the frontier model problem.
Persistent Work on Frontier Intelligence
The other half of the frontier model problem is persistence. ChatGPT and Claude are exceptional tools. But every session starts from zero. You re-upload your CSV, re-explain your business context, re-ask the same follow-up questions. When GPT-5 drops and you upgrade, you don't get smarter analysis on work you've already done — you just get a smarter starting point for the next ephemeral session.
Sourcetable's spreadsheet platform is persistent. Your Salesforce data connects once and stays connected. Your financial models auto-update from live APIs. Your dashboards pull fresh data every morning. And as the frontier advances, all of that persistent work gets the benefit automatically. You don't have to redo anything to take advantage of a better model — it's just there, working on your real data, every day.
Stop Thinking About the Model
Sourcetable's frontier guarantee:
- ✅ Always GPT-5 today — always the frontier, forever
- ✅ Automatic upgrades — no action required when a better model drops
- ✅ 100% on Vals.ai finance benchmark — frontier performance, verified
- ✅ 100+ data connectors — your analysis runs on real data, not chat sessions
- ✅ Persistent work — past analysis gets smarter as the frontier advances
- ✅ Natural language queries against Postgres, MySQL, Salesforce, and more
- ✅ One subscription — instead of ChatGPT + Claude + Gemini
