Status

Model status

Checked from our own monitor every 5 minutes. All times UTC.

All models operational

2026-11-04 02:52 UTC

GLM-5.3 Flash Abliterated

Operational

100% canary success, 90 days

90 days agoToday (UTC)

Last canary: passed 2 min ago
Time to first token: 491 ms
Latency: 503 ms
Daily model check: No daily check yet

14 of 14 canaries passed in the last 90 days.

DeepSeek V4.1 Flash

Official DeepSeek APIOperational

100% canary success, 90 days

90 days agoToday (UTC)

Last canary: passed 2 min ago
Time to first token: 823 ms
Latency: 910 ms
Daily model check: Not applicable (official API)

14 of 14 canaries passed in the last 90 days.

DeepSeek V4 Pro

Official DeepSeek APIOperational

100% canary success, 90 days

90 days agoToday (UTC)

Last canary: passed 2 min ago
Time to first token: 980 ms
Latency: 1.18 s
Daily model check: Not applicable (official API)

14 of 14 canaries passed in the last 90 days.

How we check

Canary, every 5 minutes

A short, unbilled request to each model asking it to reply with one word. We record whether it answered, the time to first token and the total latency. Each bar above is one UTC day of canaries; hover a bar for its success rate.

Daily model check, 00:05 UTC

Five benign prompts that filtered models often refuse: fiction with a villain, how everyday locks work, harm-reduction questions, a profane rap roast and a comedic roast. We count answers containing a refusal phrase such as “I can't” or “As an AI”. It doesn't run on Official DeepSeek API models: they are DeepSeek's stock models, and the check exists to show that our abliterated models answer.