182 AI models. One API key.
Chat, image, video, audio. OpenAI/Anthropic/Gemini compatible.
62 direct-from-lab routes. Price + success rate on every model.
Picking a DeepSeek route is not about choosing “the best” — it’s about choosing what you actually need.
Across eight options:
· Highest success rate (AR, 99.5%) costs ¥1.56/1M tokens
· Most expensive route (TM-deepseek-officially) costs ¥4.992 but only hits 99%
· Cheapest (XC-China, ¥1.1388) delivers 97.2% success
You’re not paying for extra reliability at 3.4x the price. You’re paying for precision that doesn’t move production metrics.
Grok Video isn’t about swapping models, it’s about picking the right route.
Two official upstream routes (MC-GK1.5 and XT-Grok): full metrics public.
TA route shows no success rate — sample size still building.
For video generation: XT-Grok is the only option with complete data + visible latency. Price difference negligible when both cost ¥0.156/sec.
Same video model, four routes, success rates ranging from N/A to 91.9%.
Page data:
· TA — Success rate N/A, HD 720p ¥0.2302/sec
· SC-Default — Success rate 83.7%, HD 720p ¥0.1248/sec
· MC-GK1.5 Official Relay Group — Success rate 89.5%, HD 720p ¥0.156/sec
· XT-Grok — Success rate 91.9%, HD 720p ¥0.156/sec, response time 58.0s
Cheapest route has no success data. Most stable route also has visible latency. 🧵