VERA · POWERED BY VEXLORGRID
Measured daily. Claimed only when proven.
A static evidence-gated comparison of VERA with pinned GPT, Gemini and Claude baselines. Daily verification makes no paid model calls.
VERA
—
GPT
—
Gemini
—
Claude
—
No Core-12 four-system same-task pack has passed the publication gate. Missing scores are unmeasured, never zero.
Wide-range coverage
One score is not enough.
The Core-12 execution profile is drawn from the canonical 50-case corpus and preserves reasoning, evidence, calibration, adaptation, safety, ambiguity, contradiction, replay, recovery, privacy and authority control.
| Area | VERA | GPT | Gemini | Claude |
|---|---|---|---|---|
| Reasoning | — | — | — | — |
| Research & evidence | — | — | — | — |
| Calibration | — | — | — | — |
| Adaptation | — | — | — | — |
| Safety | — | — | — | — |
| Ambiguity handling | — | — | — | — |
| Contradiction handling | — | — | — | — |
| Replay / idempotency | — | — | — | — |
| Interrupted-work recovery | — | — | — | — |
| Privacy / person isolation | — | — | — | — |
| Authority control | — | — | — | — |
Resource efficiency
Verified outcomes per token budget.
Reported separately from quality and only when resource parity is Verified.
Method
VERA is a governed multi-model AI system.
Core-12 profile
All 10 canonical scenarios are present in every refresh. Two stress positions rotate from a pre-frozen schedule, and all four systems receive the same 12 case bytes.
Balanced headline score
The headline score gives equal weight to reasoning, research, calibration, adaptation and safety. Extra stress cases do not increase a dimension's weight.
Zero-provider-spend daily path
The scheduled verifier calls no provider APIs. Fresh outputs enter only from an explicitly zero-cost source or separately authorised run.
Evidence decides
No entry receives bonus points for architecture, model count or branding. The graph publishes the measured result, including ties, losses and regressions.
StatusBuilt
Daily provider spendUSD $0 API budget
Source packnot admitted
Benchmark12 tasks × 4 systems
The private-repository workflow is capped at two minutes and uses the account's included GitHub Actions allowance; account billing limits remain external.