RunInfra Coding Plan
GlobalRunInfra is an open-weight inference platform launched 2026-10-20 whose Coding Plans replace per-token billing with a flat monthly fee: each tier maps to a monthly usage allowance (roughly 1.2x the subscription price in dollar usage), metered on 5-hour and weekly windows, with optional continued spend from credits or a free slower Standby mode at the limit. Its session-aware routing pins repeated context from one agent session to the same replica — the vendor claims cache hit rates up to 99.5%, stretching each plan further. Eight open-weight models are hosted (GLM 5.3 / 5.3 Flash, DeepSeek V4.1 Flash and the 1017 build, Qwen3.8 27B, Ornith 1.5 122B/35B, Nemotron 3.5 Lightning 30B) behind one API key that speaks both the OpenAI and Anthropic protocols, verified in Claude Code, Codex, OpenCode, Cline, Droid and Grok Build. Zero data retention, no training on prompts, SOC 2 Type II. Separately offered: pay-as-you-go Model APIs (from a $5 top-up) and Enterprise (dedicated infrastructure, SLAs up to 99.99%).
Verified 2026-10-24
RunInfra Coding Plan Plans
Starter（月付）
- ✓$10/month including about $12 of model usage
- ✓Free Standby: up to $0.20 of slow chat per day
- ✓4 to 32 concurrent requests per model
- ✓Session-aware routing pins repeated context to one replica — up to 99.5% cache hit rate claimed; cached input from $0.01 per 1M tokens
- ✓Works in Claude Code, Codex, OpenCode, Cline, Droid and Grok Build — one prompt to set up
- ✓At the limit: buy credits to continue, or use free slower Standby
- ✓Allowance is shared by the whole workspace, not per member
Pro（月付）
- ✓$29/month including about $35 of model usage
- ✓Free Standby: up to $0.58 of slow chat per day
- ✓5 to 32 concurrent requests per model
- ✓Session-aware routing pins repeated context to one replica — up to 99.5% cache hit rate claimed; cached input from $0.01 per 1M tokens
- ✓Works in Claude Code, Codex, OpenCode, Cline, Droid and Grok Build — one prompt to set up
- ✓At the limit: buy credits to continue, or use free slower Standby
- ✓Allowance is shared by the whole workspace, not per member
Team（月付）
- ✓$99/month including about $119 of model usage
- ✓Free Standby: up to $1.98 of slow chat per day
- ✓8 to 32 concurrent requests per model
- ✓Session-aware routing pins repeated context to one replica — up to 99.5% cache hit rate claimed; cached input from $0.01 per 1M tokens
- ✓Works in Claude Code, Codex, OpenCode, Cline, Droid and Grok Build — one prompt to set up
- ✓At the limit: buy credits to continue, or use free slower Standby
- ✓Allowance is shared by the whole workspace, not per member