EDITORIAL STANDARDS, CADENCE MATH & INDEPENDENCE
How ModelRumors.ai Works
ModelRumors.ai is a free buyer's guide for developers and technical leads choosing which frontier AI model to adopt today and when to wait for the next release. Here is how we verify data, how verdicts and forecasts are calculated, what we do not know, and how sponsorship is kept separate from editorial decisions.
1. SOURCING & DATA ACCURACY
Every Number Has a Clickable Source
Every factual claim on ModelRumors.ai cites a primary source a reader can click and verify:
- Release dates, context windows, and official prices come from each lab's own documentation, model cards, release notes, or official blog announcements (OpenAI, Anthropic, Google DeepMind, xAI, Meta, and DeepSeek).
- Live multi-provider API pricing on the Price Tracker comes from OpenRouter's public
/api/v1/modelsendpoint, polled every 6 hours and appended once per day todata/price_history.json. - Benchmark scores cite official model cards or established independent leaderboards: LMArena (arena.ai) for human preference Elo, Epoch AI for GPQA Diamond, and Artificial Analysis for output speed and intelligence index measurements.
- Missing numbers stay blank. If a lab has not published a benchmark score for a current model (for example, no frontier lab publishes SWE-bench Verified across its current late-2026 lineup), we write "Not published" rather than estimating or copying unverified aggregator pages.
2. BUYER'S GUIDE VERDICTS
How "Adopt Now", "Mid-Cycle", and "Past Typical Gap" Work
Each lab receives one of three plain-English verdicts computed directly from the days elapsed since its last public release compared to the median of its last 4 public release gaps:
Days since the last public release is less than half (50%) of the lab's typical (median) recent gap, giving teams a fresh runway.
Days since the last public release is between 50% and 100% of the lab's typical recent gap. Safe for API adoption, with an eye on the upcoming release window.
Days elapsed has reached or exceeded 100% of the lab's typical recent gap. Avoid locking in long-term fine-tunes or reserved GPU clusters until the next release lands.
3. FORECASTS & WHAT WE DO NOT KNOW
Plain-Language Cadence Math and Its Limits
Our release forecasts use simple, transparent arithmetic: we list the verified dates of a lab's 2026 public releases (API releases, public API previews, and open-weight drops), compute the gaps in days between consecutive releases, and take the range (shortest to longest) and median of the last 4 gaps. Every forecast is logged in our public Track Record so readers can judge our accuracy over time.
What we do not know: Past release cadence is a baseline, not a schedule. We do not have private visibility into internal training runs, safety-testing holds, export-control reviews, or sudden roadmap shifts inside any lab. We never attach invented statistical confidence percentages to release forecasts, and any press report about an upcoming launch is explicitly labeled: "Rumor, reported by <outlet> on <date>" with a link to the article.
4. EDITORIAL INDEPENDENCE
How Sponsorship Is Kept Separate From Verdicts
ModelRumors.ai is funded through clearly labeled sponsorships on our Advertise page (such as a "Sponsored" host row under a model card or a single sponsor line in Drop Alert emails) and referral links. Sponsors cannot buy, edit, delay, or influence an Adopt/Hold verdict, a benchmark table, or a release forecast.