The result concerns the tested linguistic construction and model snapshot, not arbitrary-predicate generalization or an identified internal mechanism.
Frozen snapshot study · September 23, 2026
What changed from GPT-3.5 to GPT-4?
Identical semantic requests. Identical settings. GPT-4 returned zero-visible-byte normal-stop responses on 30/30 null trials. GPT-3.5 returned visible text on all 30. Both models returned visible text on all matched controls.
V0: successful response, exactly empty text, normal stop. A pinned-snapshot comparison, not a same-date emergence study or a claim about the internal stopping mechanism.
SwiftAPI VOID · Runtime release status
Verify before continuation.
Inspect release availability and the published native-continuation research.
A frozen Qwen3-4B model supplies the answer. A learned 101-parameter gate changes only the native termination score, controlling whether that answer enters generation.
The answer score stays at 53.0 in every arm. Only the native EOS score moves.
First-token logits. Native EOS above the answer line terminates generation; below it, the unchanged answer is emitted. This chart presents saved model outputs, not a live simulation.
0 non-EOS logits changed.The complete non-EOS first-token vector is byte-identical across all five arms.
A separate finite-precision certificate covers every accepted six-digit equality condition across the 188 frozen bank computations. Inspect the full-domain certificate →
Research progression
Observation. Intervention. Separation.
Void measurements
Observe noncontinuation.
Condition-sensitive zero-visible-byte executions in frontier models.
A provider-aware classification of successful model executions with zero visible output bytes.
A Void is a model execution returning a successful provider response with exactly zero visible UTF-8 output bytes. Provider termination metadata determines its subtype. Explicit refusals, safety blocks, tool-mediated executions, and transport, protocol, billing, quota, rate-limit, and infrastructure failures are distinct non-Void outcomes.
VOID is not an error label. It is a precisely classified successful execution with zero visible output bytes, separated from refusals, safety events, tool calls, and operational failures.
Cross-vendor evidence
A frozen record across providers, models, and matched controls.
A frozen cross-vendor behavioral study of successful zero-visible-byte executions across 11 exact model identifiers and four providers.
A frozen GPT-5.4 study documenting a reproducible prompt-conditioned Arabic-Hebrew hybrid Unicode output regime under two system prompts differing by one Hebrew code point.
This is an infrastructure specification, not an empirical model-behavior study.
From research to infrastructure
Operational evidence, not the headline.
The laboratory's infrastructure exists to preserve, verify, and operationalize research findings. It is supporting evidence that the work can move from publication into real systems.
Signed evidence
Ed25519 execution attestations with offline verification.
Revocation and replay defense
Immediate revocation checks, unique identifiers, and expiry.
Policy and authority
Scoped keys, policy-deny behavior, grants, and audit records.
Integration surfaces
SDK, API, proxy, and runtime enforcement interfaces.
Evaluate behavior through your real deployment path.
SwiftAPI Labs conducts paid, time-bounded evaluations for organizations deploying frontier models and autonomous agents. Evaluations reproduce model behavior through the customer's route, preserve raw provider metadata, classify terminal states, map retry and continuation behavior, and produce a written implementation decision.
