On February 10, 2026, between 8:03 AM and 9:48 AM PST, paid ChatGPT plans experienced elevated error rates due to a temporary serving capacity shortfall in one of our GPT-5.2 model fleets. Demand exceeded available capacity, resulting in conversation failures. We mitigated the issue by enabling load shedding, reallocating compatible capacity, and bringing additional serving engines online. A brief secondary impact occurred during mitigation due to incompatible capacity being temporarily assigned and then removed. Service was stabilized by 9:48 AM PST.
Paid ChatGPT plans experienced elevated conversation error rates.
Impact duration was approximately 105 minutes.
A brief secondary impact affected ChatGPT Go due to related capacity adjustments.
The incident was caused by a reduction in effective serving capacity for one of our paid GPT-5.2 model fleets. Capacity had not fully recovered following a prior engine reallocation, leaving the system operating with less headroom than expected.
On the morning of February 10, traffic demand exceeded the available capacity, leading to overload conditions and increased request failures.
During mitigation, some temporary capacity shifts between service tiers introduced configuration incompatibilities, which resulted in a small number of additional transient errors before being corrected.
We took the following actions to restore stability:
Enabled load shedding to reduce pressure on the system.
Disabled certain retry behaviors to prevent amplification under overload conditions.
Reallocated compatible serving capacity to the affected fleet.
Provisioned additional serving engines.
Removed incompatible capacity once identified.
As additional capacity came online and system pressure decreased, error rates returned to baseline levels. The incident was marked as mitigated at 9:48 AM PT.
Following this incident, we are implementing the following improvements:
Improvements to capacity reconciliation and visibility to help support serving capacity to match expected levels.
Stronger validation when moving capacity between service tiers to help prevent configuration incompatibilities.
Operational playbook enhancements to help accelerate safe capacity reallocation during incidents.
We apologize for the disruption and appreciate your patience while we worked to restore full service reliability.
