DigitalOcean replaces four preset routers with inference-router and three modes

On 12 October DigitalOcean moved its Inference platform from four presets (general, software-engineering, writing, knowledge-base-document) to a single inference-router with a routing_preference of balanced, budget or quality. Across 20 short prompts the router mostly picked Kimi K3 and cost $4.94–5.55 per 1,000 requests, more than calling Claude Opus 5.5 directly at $3.48.

• Router picked Kimi K3 for 14–15 of 20 prompts in all three modes
• Cost per 1,000 requests: $4.94 balanced, $5.04 quality, $5.55 budget
• Direct Claude Opus 5.5 cost $3.48; old router:general cost $0.31
• For the prompt "hi" the router spent $0.0034 versus $0.000008 on Haiku 5.5