Historical version
This page records design, operation, or validation facts from an earlier release. It is not current V7.17.3 guidance. Start from the documentation hub for current instructions.
V7.13.0 Release Notes¶
This compatibility-support release separates model positioning, reasoning gains and budget approval. Frozen V3 remains the default, and new models are available only for explicitly enabled evaluation. Codex Desktop is the sole product target; internal scripts provide installation, management and validation.
- Explicit V4 lists GPT-5.6 Luna/Terra/Sol and GPT-6 Luna/Sol/Astra at Low/Medium/High: 18 evaluation profiles. The nine GPT-6 production candidates require scenario qualification before activation.
- Same-model effort increases and model switches use direct paired-gain evidence. Whole phase allocations protect future reviews and repairs.
- Budget V4, Review State V9, Result V6 and Sample V4 add evaluation, publication, revocation and source-verified observation controls.
- Frozen readers retain old policies, ledgers and results. GPT-5.6 remains available through explicit compatibility paths. Missing qualification returns CALIBRATION_REQUIRED without replacing existing tasks.
- Installation separates the Desktop management component from runtime capabilities and removes the standalone CLI release matrix dependency. An internal prerelease component version is not a Desktop capability verdict.
- Normalize staged Windows launchers to CRLF so Git checkout settings cannot change release archives; source files and frozen ledgers are not rewritten.
GPT-6 production defaults are not switched in this version. Native V4 admission remains unaccepted: configuration reports trust, but both an existing task and an independent fresh task lacked budget reservations and creation receipts. The current environment remains policy-constrained; enforced-budget effectiveness is not claimed. Real model qualification, effort/model-switch gains and default activation require later acceptance, not installation, trust status or synthetic tests. See validation and review for code/artifact evidence and known limits. Public publication is established separately by the Git tag, release assets and provenance readback.