Funding

Funding sources are public and prominent because independence is what makes the record credible.

Hard rules

  • We do not accept funding from any commercial LLM provider, now or ever.
  • We do not accept paid placement, sponsored content, or preferential treatment in analysis.
  • No funder sees a report before it is published.
  • All funding sources are listed on this page.

Current sources

The project is bootstrapped from volunteer labor and out-of-pocket API credits. We accept:

  • Individual donations via Buy Me a Coffee. Donors are listed by BMC unless anonymous. No editorial influence attaches to a donation — we do not know who has donated when writing reports, and would not change a report’s content if we did.
  • Grants from AI-safety and civil-society foundations, once we apply.
  • Patronage from journalism or research organizations whose work benefits from the measurements.
  • Institutional subscriptions for deep-access to raw data, while the aggregate dashboard stays free.

Buy me a coffee

Detailed funding disclosures will be posted here once there is anything beyond volunteer labor to disclose.

Cost transparency

The dominant expense is API costs at the providers we measure. Current monthly burn is ~$86 on the Level 0 configuration: 30 prompts, Claude Opus and GPT-5 Preview alternating biweekly, Ollama baseline every week. That’s volunteer-sustainable and is what we’re running today.

Infrastructure (GitHub Pages, storage, worker compute) is currently absorbed by the maintainer.

What funding unlocks

The project started deliberately narrow. We want to broaden. Each step up requires sustained monthly funding; sponsors move the project up this ladder.

Now · ~$86 / mo
30 prompts. Opus and GPT-5 Preview alternating biweekly. Ollama every week. Current volunteer-sustained operation.
+$150 / mo
Add Google Gemini 2.5 Pro on a biweekly cadence. First time we cover all three major providers weekly at the frontier tier.
+$300 / mo
Unalternate: Opus, GPT-5, and Gemini all run every week. Weekly time-resolution on every frontier model.
+$800 / mo
Expand the corpus from 30 to ~100 prompts. Hits the coverage the methodology calls for and makes per-axis analysis statistically robust.
+$1,500 / mo
Full methodology-spec corpus (150–200 prompts), full roster weekly, held-out set fully populated, plus durable storage (S3 + IPFS pinning for the “retention forever” guarantee).
+$2,500 / mo
200–300 prompt corpus at full sampling density. Trained refusal classifier. Postgres index for researcher queries. Scheduled orchestration. Upper end of what the project can productively spend.

Buy me a coffee

One coffee (~$5) funds roughly a week of the current Level 0 configuration, or about an hour of measurement on the frontier tier at Level 1.