Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
GPT-5.6 Sol Repeatedly Hits “Selected Model Is at Capacity” in Codex Desktop: Causes, Fixes, and Current Status
If you’ve been trying to use GPT-5.6 Sol in Codex Desktop only to encounter the message:
“Selected model is at capacity. Please try a different model.”
or receive server_overloaded errors, you’re not alone.
Several users have reported intermittent capacity issues with GPT-5.6 Sol, particularly within Codex Desktop, even when other models continue to function normally. Interestingly, many users have also noted that the same model continued working from the terminal (CLI) while failing inside the desktop application.
In this guide, we’ll explain why this happens, what you can do, and what recent user reports suggest.
The Error
Users commonly see one of the following messages:
Selected model is at capacity.
Please try a different model.or
server_overloadedThe problem may occur:
- Immediately after selecting GPT-5.6 Sol
- During conversation generation
- While switching between models
- Even after restarting Codex Desktop
Network Issues Are Often Not the Cause
One user reproduced the issue after testing multiple scenarios:
- Different internet connection
- Mobile hotspot
- VPN disabled
- Secure DNS disabled
The problem still occurred.
This strongly suggests that, in many cases, the issue originates from the server side rather than local networking.
OpenAI Previously Acknowledged Similar Issues
OpenAI’s status page previously reported an incident similar to this:
Codex 5.6-sol Experiencing Increased Server-Overload Errors
The incident was marked as resolved.
However, some users later reported seeing the exact same behavior again several days afterward.
This indicates that server capacity problems can occasionally reappear during periods of heavy demand.
Interesting Observation: Terminal Continued Working
A particularly interesting finding reported by users:
- Codex Desktop failed
- GPT-5.6 Sol continued working from the terminal/CLI
This may indicate:
- Different request routing
- Separate infrastructure paths
- Different retry behavior
- Different queue management
Although OpenAI hasn’t officially confirmed these implementation details, the behavior has been observed by multiple users.
Common Symptoms
Users generally report that:
- GPT-5.6 Sol fails
- Other GPT models work normally
- Desktop application shows capacity errors
- Terminal version works
- Restarting the application doesn’t help
- Network changes make no difference
Possible Causes
Several factors may contribute to these errors.
1. Temporary Server Capacity
The most likely cause is that GPT-5.6 Sol temporarily reaches its request limit.
Large reasoning models require significantly more compute resources than smaller models, making them more susceptible to overload during peak usage.
2. Regional Load Balancing
Requests from different geographic regions may be routed to different clusters.
As a result:
- One region may experience overload.
- Another region may continue working normally.
This explains why some users report no issues while others consistently encounter capacity errors.
3. Desktop-Specific Queueing
If the desktop application communicates with a different backend service than the CLI, temporary inconsistencies may occur.
This could explain why:
- CLI works
- Desktop doesn’t
during the same time period.
4. Ongoing Infrastructure Updates
Large AI services continuously update routing, scaling, and deployment systems.
During these rollouts, short-lived capacity problems can occasionally affect specific models.
Things You Can Try
While the problem is often server-side, the following steps are worth trying:
Restart Codex Desktop
A fresh session may reconnect to a different backend.
Switch Models
Use another available model temporarily until GPT-5.6 Sol becomes available again.
Try the Terminal Version
Some users have reported that GPT-5.6 Sol continued functioning normally through the terminal even while the desktop application showed capacity errors.
Test Another Network
Although many users found this didn’t solve the problem, testing with:
- Mobile hotspot
- Different ISP
- Different Wi-Fi
can help rule out local networking issues.
Disable VPN
VPN routing can occasionally introduce connection problems.
Testing without a VPN helps isolate the issue.
Submit Feedback
Using Codex Desktop’s /feedback option sends diagnostic information that may help OpenAI investigate recurring problems.
Check the OpenAI Status Page
Before spending time troubleshooting locally, verify whether OpenAI has reported an active incident affecting Codex or model availability.
Good News: Many Users Reported the Problem Eventually Disappeared
One encouraging update from the community came from a user who later reported:
“I was able to use every model again without problems. I suppose they fixed it.”
This suggests that many of these incidents are temporary and are resolved once backend capacity stabilizes.
Is This a Desktop Bug?
Not necessarily.
Based on user reports, the evidence points more toward intermittent server capacity than a permanent bug in Codex Desktop itself.
If:
- every other model works,
- terminal access works,
- multiple networks show the same behavior,
the issue is likely occurring upstream rather than on your computer.
Information That Helps Diagnose the Issue
If you’re experiencing the same problem, sharing the following details can help identify patterns:
- Operating system (Windows, macOS, Linux)
- Codex Desktop version
- Date and time (including time zone)
- Internet provider or network type
- Whether another network shows the same behavior
- Whether the terminal/CLI works
- Whether other GPT models function normally
- Exact error message displayed
The more reports available, the easier it becomes to determine whether the issue is regional, platform-specific, or part of a broader service disruption.
Final Thoughts
Intermittent “Selected model is at capacity” and server_overloaded errors for GPT-5.6 Sol can be frustrating, especially when every other model appears to work normally. Current community reports suggest these incidents are usually related to temporary server-side capacity rather than problems with your internet connection or device.
If you’ve already tested different networks, disabled VPNs, and confirmed that other models work, the best approach is often to use an alternative model temporarily, monitor OpenAI’s service status, and report the issue through the in-app feedback tool if it persists.
Frequently Asked Questions (FAQs)
Why does GPT-5.6 Sol say “Selected model is at capacity”?
It usually means the model is temporarily unable to accept new requests because demand exceeds available server capacity.
Does changing my internet connection fix the problem?
In many reported cases, switching to another network, including a mobile hotspot, did not resolve the issue, indicating a server-side cause.
Why does GPT-5.6 Sol work in the terminal but not Codex Desktop?
Some users have observed this behavior, possibly due to differences in backend routing or request handling between the CLI and desktop application.
Is this a permanent issue?
No. Many users report that the problem disappears after OpenAI restores capacity or resolves backend issues.
Should I reinstall Codex Desktop?
Reinstalling is unlikely to help if the issue is caused by temporary server overload, though restarting the application is worth trying before assuming it’s a service-side problem.
