ModelSheet
A decision tool for non-technical founders and operations leads that audits which AI models a business is currently paying for, scores each against use case, data sensitivity, and budget, and recommends an open-weights vs proprietary mix tailored to the company's reality.
operations leads and founders at sub-500-person companies picking or renegotiating AI vendors
- Vendor stack audit that ingests invoices, API usage logs, and chat transcripts to show what you actually pay per model per workflow
- Open vs closed scoring matrix that weighs data residency, switching cost, performance on your own prompts, and geopolitical risk
- Migration playbooks that walk through moving one workflow at a time (e.g. support summarization, marketing copy) from a proprietary model to an open-weights alternative without downtime
- Quarterly re-check that flags when a new open model closes the gap on a task you're paying a closed vendor for
The HN discussion explicitly shows the closed-vs-open gap has closed and a16z partner data says 80% of startups use Chinese models — business buyers who have never thought about model strategy are now exposed to real risk and a real opportunity, and they have no neutral tool to help them decide.
a16z partner Martin Casado's 80% Chinese-model claim is verified by Economist and Business Insider, but the specific buyer — non-technical ops leads at sub-500-person firms — is unlikely to be actively shopping for a model-strategy tool; pain is more acute among technical founders.China is quietly upstaging America with its open models ↗80% Of Startups Using Open-source Models Are Using Chinese Models ↗
Existing options are either SEO/affiliate comparison blogs (lmmarketcap, aicostcheck, localalternative) or infrastructure routers (OpenRouter, Helicone); no neutral SaaS that audits a company's current AI spend and recommends an open-vs-proprietary mix appears to exist yet.Open Source vs Proprietary AI Models - LM Market Cap ↗Proprietary vs Open-Source AI Models: How to Choose in 2026 ↗
Cost-savings story ($0.14 vs $30/M tokens per a16z coverage) is compelling, but non-technical ops leads typically use free calculators, vendor sales calls, or fractional CTOs; willingness to pay for a standalone decision tool in this segment is unproven.80% of Startups Now Choose Chinese AI Models Over US Alternatives ↗
Structural drivers (cost gaps, vendor lock-in fears, data-sovereignty concerns around Chinese models) sustain the need, but the model landscape shifts every few months, requiring constant benchmark refresh and creating ongoing maintenance burden.AI race splits in two as China wages open-weight insurgency ↗Many of the Chinese models are open weights, so if you ... - Hacker News ↗
Buildable with public benchmarks (HuggingFace leaderboards, Artificial Analysis, lmmarketcap) and vendor pricing pages; main cost is keeping pricing/benchmark data fresh and building the recommendation heuristics, not the core tech.Open-Source vs Proprietary AI Models: A Complete Cost Comparison for 2026 ↗