All checks passing

Model availability

Every model you can call, worst state first. Problems are listed openly; the full list is folded below. A model verified by real customer traffic is a stronger claim than an upstream host that merely answered, so the two are labelled differently and never share a badge.

Each row states its evidence: real customer requests where they exist, upstream reachability where they do not, and inference across models sharing an upstream host. The 24-hour figure is the share of observed samples that succeeded — an observation, not a contractual uptime percentage — and a model with no samples yet shows no figure rather than a flattering one.

Last 90 days

Most-used models

The 12 public models with the most real customer traffic over the window, one cell per day, oldest on the left. A day's figure is the share of that day's health checks that found the model working; a day with too few checks shows grey, never green. Every other model's 90-day figure is in the full list above.

1. DeepSeek V4.1 Flash

   deepseek-v4.1-flash

   90d 92.56%

2. DeepSeek V4 Flash

   deepseek-v4-flash

   90d 95.32%

3. Claude Sonnet 5

   claude-sonnet-5

   90d 92.74%

4. Claude Opus 5.5

   claude-opus-5-5

   90d 86.85%

5. GPT-6 Luna

   gpt-6-luna

   90d 96.98%

6. GPT-5.6 Luna

   gpt-5.6-luna

   90d 97.69%

7. GLM 5.3 Flash v3

   glm-5.3-flash-v3

   90d 94.07%

8. Claude Opus 5

   claude-opus-5

   90d 94.69%

9. GPT-6 Sol

   gpt-6-sol

   90d 98.38%

10. Gemini 3.1 Flash Lite Preview

    gemini-3.1-flash-lite-preview

    90d 98.66%

11. GPT-6 astra

    gpt-6-astra

    90d 96.19%

12. Sonnet 4.6

    claude-sonnet-4-6

    90d 93.77%