OpenGPTEvidence-based model decisionsLMArenaOpen source
Reading live state
Automatic after validation
TRUST & DATA STATUS
What is checked, reviewed, and published
See the evidence status behind OpenGPT's model and Agent rankings, including source dates, latest checks, and publication controls.
Status policy updated
01
Model ranking sources
Live monitoring states come from the public status endpoint. Source dates remain separate from check times.
SourceEvidence stateSource data dateLast checkedPublication control
Artificial AnalysisOpen source
Reading live state
Not published by source
Human review before publication
LiveBenchOpen source
Reading live state
Human review before publication
02
Agent evaluation sources
Agent rows show the published evidence record and its latest recorded check. They do not claim source or product uptime.
SourceEvidence stateSource data dateLast checkedPublication control
SWE-bench VerifiedOpen source
Historical snapshot
Human review before publication
Terminal-Bench 2.1Open source
Published snapshot
Human review before publication
AssistantBenchOpen source
Historical snapshot
Not published by source
Human review before publication
WebArenaOpen source
Source review required
Not published by source
Human review before publication
OSWorld 2.0Open source
Published snapshot
Human review before publication
Berkeley Function-Calling LeaderboardOpen source
Historical snapshot
Human review before publication
GAIAOpen source
Historical snapshot
Not published by source
Human review before publication
τ²-bench CoreOpen source
Published snapshot
Not published by source
Human review before publication
τ³-Banking KnowledgeOpen source
Published snapshot
Human review before publication
03
Service boundaries
A status label only describes the specific workflow named here. It should not be read as a broader availability promise.
Public rankings
OpenGPT publishes source snapshots and catalog records. It does not continuously test provider APIs, model endpoints, prices, or regional access.
Source monitoring
Configured upstream artifacts are checked on their stated cadence. A successful check does not prove every source page, model, or API is available.
Local decision tools
Saved candidates, task inputs, and reports stay in this browser by default. This status page does not imply cloud sync or recovery.
Submissions and email
Submissions enter editorial review. Receipt and final-result emails use a queued transactional process; instant delivery or a response time is not promised.
04
How publication decisions work
Collection, review, catalog inclusion, comparability, and ranking remain separate decisions.
Validated automatic publication
Only LMArena is eligible, and only after schema, completeness, date, and anomaly checks pass. Validation fails closed.
Human source review
Artificial Analysis, LiveBench, and Agent sources require human review before published evidence changes.
User submissions
Approval can list a reviewed Agent product in the public directory. It does not set rank, comparability, evidence grade, or guarantee leaderboard inclusion.
05
Methods and corrections
Read the comparison rules or report an inaccurate date, mapping, link, or evidence label.