Cerebras had 27 outages in the last 2 years totaling 64h 22m of downtime — averaging 1.1 incidents per month.
There were 27 Cerebras outages since January 11, 2025 totaling 64h 22m of downtime. Each is summarised below — incident details, duration, and resolution information.
Customer using AWS billing may experience availability issues. Exact reason is being investigated.
AWS billing issue is fixed now.
The service is currently inaccessible. We are currently working urgently to restore service capabilities. We will provide further updates as we make progress.
This incident has been resolved.
Between 9:07 AM UTC and 9:27 AM UTC users experienced service disruption with qwen-3-235b-a22b-instruct-2507. We have deployed a fix and the issue is now resolved.
Between 9:07 AM UTC and 9:18 AM UTC users experienced partial service disruption with gpt-oss-120b. We have deployed a fix and the issue is now resolved.
Between 10:30 PM PST and 11:42 PM PST 03/17 users experienced partial degradation with glm-4.7. We have deployed a fix and the issue is now resolved.
The service is currently inaccessible. We are currently working urgently to restore service capabilities. We will provide further updates as we make progress.
This incident has been resolved.
We are currently investigating an issue with zai-glm-4.7 where users are seeing an elevated number of 503 errors. Engineering is working on a resolution.
We've identified a fix and our engineering team is deploying this now.
This incident has been resolved.
We are currently investigating an issue with Llama-3.3-70B where users are seeing an elevated number of 503 errors. Engineering is working on a resolution.
We've identified a fix and our engineering team is deploying this now.
Our engineering team has resolved the issue.
This incident has been resolved.
We are investigating 502 Gateway errors on Cerebras endpoints. We are currently working to resume normal service performance. We will provide further updates as we make progress.
Between 7.30 PM PT and 11 PM PT, the inference service was partially disrupted by 502 Gateway errors across all endpoints. This issue is caused by an internal system dependency, the fix was rolled out and the service is operational across all endpoints.
Qwen 235B performance is degraded and temporarily unavailable for some service tiers. We are currently working to resume normal service performance. We will provide further updates as we make progress.
As part of recent changes in traffic patterns and capacity, we are temporarily turning down rate limits for Pay-go users to help maintain a positive experience across the board. We understand the challenges this may create for you and your users, and we sincerely apologize for the inconvenience. We plan to monitor this endpoint and evaluate restoring rate limits in the next week. Thank you for your continued partnership and understanding, please reach out to our support team for any other questions.
Between 06:00 AM PT and 06:00 PM PT users experienced service disruption with Qwen 3 235B Instruct. We have taken action to address recent changes in traffic patterns and capacity, reducing the disruption. We plan to monitor this endpoint and evaluate restoring rate limits in the next week for Pay Go users.
Between 06:00 AM PT and 06:00 PM PT users experienced service disruption with Qwen 3 235B Instruct. We have taken action to address recent changes in traffic patterns and capacity, reducing the disruption. We plan to monitor this endpoint and evaluate restoring rate limits in the next week for Pay Go users.
Qwen 32b is currently inaccessible. We are currently working urgently to restore service capabilities. We will provide further updates as we make progress.
Partial of the service has been restored, and we are currently working to resume normal service performance
This incident has been resolved.
The service is currently inaccessible. We are currently working urgently to restore service capabilities. We will provide further updates as we make progress.
The issue has been root caused and fix is being implemented to bring the service backup.
The fix has been rolled out and the service has resumed consuming traffic and being monitored.
Between 12.45 PM PT and 4.00 PM PT users experienced service unavailability with Llama 3.3 70B, caused by a datacenter issue. We have deployed a fix and the issue is now resolved, model endpoint is operational.
Between 12.45 PM PT and 4.00 PM PT users experienced service unavailability with Llama 3.3 70B, caused by a datacenter issue. We have deployed a fix and the issue is now resolved, model endpoint is operational.
Partial service disruption due to 401 Error with Unauthorized API Key. We are currently working to resume normal service performance. We will provide further updates as we make progress.
This issue is caused by an internal system dependency, and we are currently working to restore system performance.
Between 4.25 PM PT and 5.45 PM PT, developers experienced minor service disruption due to API Key Error. The issue has been identified and resolved, normal service operation is restored.
We are currently working to resume normal service performance. We will provide further updates as we make progress.
We are continuing to investigate this issue.
We are continuing to investigate this issue.
We've deployed the fix and affected service is now recovering. We are actively monitoring service performance.
Between 06:50 AM PST and 07:45 AM PST users experienced partial degradation with llama3.1-8b, llama-3.3-70b, qwen-3-32b, qwen-3-235b-instruct-2507 models. We have deployed a fix and the issue is now resolved.
Between 06:50 AM PST and 07:45 AM PST users experienced partial degradation with llama3.1-8b, llama-3.3-70b, qwen-3-32b, qwen-3-235b-instruct-2507 models. We have deployed a fix and the issue is now resolved.
Between 06:50 AM PST and 07:45 AM PST users experienced partial degradation with llama3.1-8b, llama-3.3-70b, qwen-3-32b, qwen-3-235b-instruct-2507 models. We have deployed a fix and the issue is now resolved.
Between 01:30 AM PT and 2:30 PM PT users experienced partial degradation with Qwen 3 235B Instruct. We have deployed a fix and the issue is now resolved.
Around 4:09 AM PT, Inference Platform went down causing 500 Internal Server Errors. This issue is caused by a third-party system, and we are currently monitoring.
We are continuing to investigate this issue.
We are continuing to investigate this issue.
The issue has been identified and a fix is being implemented.
A fix has been implemented and we are monitoring the results.
The platform is accessible now. We are monitoring to ensure stability.
This incident has been resolved.
Between 1:30 AM PST and 2:30 AM PST users experienced service disruption with ZAI-GLM-4.6. We have deployed a fix and the issue is now resolved.
We are experiencing degradation of service across several of our models. Our engineering team is working to identify the root cause and restore stability to our systems.
We've identified that the combination of high load and a network routing issue resulted in higher queue times and rejection rates. ETA on resolution is a few hours.
Our teams are continuing to work to resolve this issue. We appreciate your patience.
A fix has been implemented, availability is returning to 100%, and latency times are improving. We will continue to monitor over the next few hours.
This incident has been resolved.
We are experiencing issues with one of our datacenters. We are actively working on mitigation. Seeing minimal degradation of service at this point. Fortunately, this only impacts one datacenter, the rest are operational, and redundancies are in place. ETA recovery time for datacenter: 8 hours. We will actively keep you updated on the progress.
We are continuing to investigate this issue.
Update: The issue has been root caused to a city water supply disruption. We are actively working on mitigation.
Between Oct 18th 7 PM PT and Oct 19th 6.30 AM PT users experienced minor service disruption with Llama 3.3 70B, Qwen 3 32B, GPT OSS 120B, Qwen 3 235B Thinking, Qwen 3 Coder 480B because of a datacenter issue. We have deployed the fix and the issue is now resolved.
Llama-3.3-70B endpoint experienced reduced availability starting at approximately 7:45am PT. Endpoint went back to full availability starting 8:55am PT.
We observed a temporary increase in latency and intermittent 503 errors for the Llama-3.3-70B model earlier today (between 10:59am Pacific to 11:09am Pacific on 7/31/2025). This was due to ongoing infrastructure operations, including system scaling and maintenance, which briefly reduced available capacity. Service availability and performance are now stabilized.
Llama-3.3-70B, DeepSeek-R1-Distill-Llama-70B, Qwen-3-32B (7/25/2025 from 1:17 PM to 1:41 PM PT) were unavailable. We restored normal service performance at 1:41 PM PT on July 25th.
We had a brief issue this morning (6/25/2025 from 7:25 AM to 9:40 AM PT) affecting the Qwen-3-32B shared endpoint. Some users may have seen 404 or 503 errors during that time.
Llama3.1-8B was unavailable. We restored normal service performance at 3:21 PM EST on June 19th.
This issue is caused by an internal system dependency, and we are currently working to restore system performance.
Some features may be temporarily unavailable. We are currently working to resume normal service performance. We will provide further updates as we make progress.
Between 04/30 23:20 UTC and 05/01 01:35 UTC, users experienced partial degradation with Llama-3.3-70B. We have deployed a fix, and the issue is now resolved.