About Jev Tracker
How we check claims.
Understand the sources, checks and limits behind Jev Tracker coverage before relying on a conclusion.
About this publication
Jev Tracker is an independent publication for people following Jev. We explain published experiments, access options, performance claims and capability limits in plain language.
We are not TypeSafe AI and have no affiliation, partnership or endorsement from the model provider. This site is not a Jev chat service, account portal or developer tool.
Contact us with questions, source suggestions or corrections through our Contact page.
Who said it matters
Source labels distinguish provider announcements, integration documentation and outside reporting. They identify who published a claim, rather than rate its accuracy.
- Official
Material from the model provider, including announcements and documentation.
Useful for what the provider states. Not an independent measurement.
- Integration
A platform or framework documenting how it connects to Jev.
Describes that integration. It does not prove every account can use it.
- Independent
Analysis published outside the model provider.
Check the author’s methods, incentives and evidence. Outside authorship alone is not reproducibility.
- Self-reported
An author’s account of their own experience or result.
Conditions may be incomplete. We do not present it as a result reproduced by Jev Tracker.
- Community
Discussion, synthesis or aggregated reports from a wider group.
Selection, missing context and incompatible tasks can make comparisons misleading.
Checked is not reproduced
Current coverage is document-based. Jev Tracker has not independently run the featured experiments or successfully called Jev to verify access.
Inspect the original material
We read the original source, identify its publisher and link to it so you can inspect the claim yourself.
Keep the conditions attached
We include the task, model, comparison baseline and relevant limitations. Peak results are labeled separately from typical outcomes.
Separate evidence from interpretation
We distinguish published findings from our interpretation and highlight questions the available evidence leaves unanswered.
Publish a dated record
We review and update our coverage manually. Each source-check date records when we last reviewed the cited material.
Three different evidence thresholds
- Source inspected
- The cited material was reviewed. This is what our source-check timestamps describe.
- Access tested
- Would require a successful request under stated account and model conditions. Reading availability documentation is not this test.
- Result reproduced
- Would require a documented run with inputs, setup and results. A linked demonstration is not a Jev Tracker reproduction.
Read the date, not just the claim
- Published
- When the source was published, where a date is available.
- Event date
- When the announcement, release or other reported event happened.
- Source checked
- When we last reviewed the cited material, shown in UTC.
Stale, partial or unavailable information
Access entries become Stale after more than 72 hours without a new source check. Refer to each entry’s date and model version when using the information.
We label unknown prices or requirements as Not yet verified, explain incomplete evidence and flag sources that are no longer available.
Keep changes in the open
For material corrections, we record the affected claim, previous wording, new evidence, reason for the change and date. We distinguish updates to our coverage from changes to the product.
The limitations guide records how explicit model documentation resolved our earlier uncertainty about image input. The note preserves what changed in our coverage.
Read a coverage update