30/32 models healthy · 4 zero-retention · availability measured from rolling probes
| model | state | availability | avg latency | in / out per 1M |
|---|---|---|---|---|
| GPT-6 Astra | 90% | 59.6s | $3.13 / $12.5 | |
| GPT-6.1 Sol | 100% | 3.4s | $3.13 / $12.5 | |
| GPT-6 Sol | 100% | 9.6s | $3.13 / $12.5 | |
| GPT-6 Luna | 100% | 2.6s | $3.13 / $12.5 | |
| GPT-5.6 Sol | 100% | 14.8s | $1.56 / $6.25 | |
| GPT-5.6 Terra | 100% | 5.2s | $1.56 / $6.25 | |
| GPT-5.6 Luna | 100% | 1.3s | $1.56 / $6.25 | |
| GPT-5.5 | 100% | 2.4s | $0.88 / $3.5 | |
| Claude Fable 5 | 100% | 7.7s | $3.13 / $15.63 | |
| Claude Opus 5.5 | 100% | 4.9s | $3.13 / $15.63 | |
| Claude Opus 5 | 100% | 5.2s | $2.5 / $12.5 | |
| Claude Sonnet 5.5 | 100% | 16.3s | $1.56 / $7.81 | |
| Claude Sonnet 5 | 100% | 4.9s | $1.56 / $7.81 | |
| Claude Opus 4.8 | 100% | 5.5s | $2.5 / $12.5 | |
| Claude Opus 4.7 | 100% | 4s | $2.5 / $12.5 | |
| Claude Opus 4.6 | 100% | 5.1s | $2.5 / $12.5 | |
| GLM-5.3 | 100% | 7.9s | $0.75 / $3 | |
| GLM-5.3 Uncensoredzdr | 100% | 7.9s | $0.9 / $3.6 | |
| GLM-5.3 Flash | 100% | 8.2s | $0.19 / $0.75 | |
| GLM-5.3 Flash Uncensoredzdr | 100% | 8.2s | $0.23 / $0.94 | |
| GLM-5.3 Flash Abliteratedzdr | 95% | 173.5s | $0.4 / $1.6 | |
| GLM-5.2 | 100% | 4.3s | $0.56 / $2.25 | |
| DeepSeek V4.1 Flash | 100% | 3.6s | $0.16 / $0.63 | |
| DeepSeek V4 Pro Uncensoredzdr | 100% | 4.2s | $0.56 / $2.25 | |
| DeepSeek V4 Pro | 100% | 4.2s | $0.56 / $2.25 | |
| DeepSeek V4 Flash | 100% | 4.2s | $0.16 / $0.63 | |
| Grok 4.7 | 100% | 19.3s | $1.56 / $7.81 | |
| Grok 4.6 | 100% | 26.7s | $1.56 / $7.81 | |
| Grok 4.5 | 100% | 20.4s | $1.25 / $6.25 | |
| Kimi K3 | 100% | 6.1s | $0.68 / $2.7 | |
| MiMo v2.6 Pro | 0% | — | $0.56 / $2.25 | |
| MiMo v2.6 Flash | 0% | — | $0.11 / $0.45 |
latency reflects model response time, not our api overhead — requests via unjailed add ~0.3s. incidents on individual models don't affect others.