Skip to content

Coverage

Outcome Result
Pinned author additions and owned telemetry reconstructed, built, independently reviewed
Tools, JSON, tool continuation, streamed tools, image passed native preflight
C4 and telemetry 4/4 exact canaries; 1164 coherent samples
Context/cache 19,456 and 99,284 token retrieval; warm and restart hits passed
Logprobs 128 accepted, 129 refused; requested IDs distinct; repeat and concurrent probes passed
Truncated Messages tool call baseline500; candidate200 parsed partial input and max_tokens
Restoration exact incumbent restored/readmitted/routed smoke passed before second delta window
Performance not collected as matched comparison; no speed claim
Video, L2PF, think-off splitting separate/off; not qualified
Fleet rollout closure native owner unavailable; separately tracked