# runaii cloud

> runaii.cloud is a serverless LLM inference platform for open models. One
> OpenAI-compatible API (https://api.runaii.cloud/v1), per-token
> pricing with cached prompt reads billed at a fraction of input, per-second GPU VMs,
> and managed training. Models are served on GPU cells runaii operates itself —
> not resold capacity.

## Key facts

- API: OpenAI-compatible `POST https://api.runaii.cloud/v1/chat/completions`
- Also supports Anthropic-format requests for Claude Code-style clients (live at https://runaii.cloud/api/messages).
- Pricing: per token. Canonical, machine-readable price list — per-model $/1M input, cached and output, per-GPU-preset $/hr, region multipliers: https://runaii.cloud/pricing.json. Cached prompt reads are billed at a fraction of input; model metadata (context, modality) is also at https://runaii.cloud/api/catalog.
- Free credits: $5 on signup (email verification required).
- GPU VMs: billed per second, from $0.30/hr (RTX PRO 6000 Blackwell 12GB slice) and
  $0.44/hr (T4) through L4 and A100/H100 VMs up to 8× B200/B300 and 16× GB300
  ($184/hr) — full ladder at https://runaii.cloud/gpus
- Coding plans ("Token-Max"): flat monthly, 100M–1B tokens per day, application-gated
  because capacity is physical — https://runaii.cloud/token-max
- Flat-rate plans ("Founding batch"): unlimited tokens within 10M/day fair use,
  $12–$99/mo ($24 tier = ₹1,999/mo in India), full served catalog on one key,
  capacity-gated — the flash node lights only when 45 seats are prepaid — https://runaii.cloud/reserve

- Cache economics, measured on our fleet: 95.8% prefix-cache hit → $0.023/M effective
  input, warm TTFT 0.28-0.93s at 1M context. Method published: https://runaii.cloud/benchmarks/cache

## Models served

- [GLM-5.3](https://runaii.cloud/models/glm-5.3) (`glm-5.3`) — 203K context, $1.4/1M input, $0.26/1M cached, $4.4/1M output
- [DeepSeek V4.1 Flash](https://runaii.cloud/models/deepseek-v4-flash) (`deepseek-v4-flash`) — 1049K context, $0.22/1M input, $0.022/1M cached, $0.66/1M output
- [Qwen3.8-27B Flash](https://runaii.cloud/models/qwen3.8-flash) (`qwen3.8-flash`) — 33K context, $0.4/1M input, $0.04/1M cached, $1.6/1M output

Live catalog (always current): [model catalog](https://runaii.cloud/models)

## Key pages

- [Homepage](https://runaii.cloud/)
- [Pricing (per-token, per-model)](https://runaii.cloud/pricing)
- [Model catalog with context and pricing](https://runaii.cloud/models)
- [GPU VMs, per-second billing](https://runaii.cloud/gpus)
- [RTX PRO 6000 Blackwell slices, 12-48GB from $0.30/hr](https://runaii.cloud/gpus/g4-slices)
- [Token-Max coding plans (apply)](https://runaii.cloud/token-max)
- [Flat-rate plans — founding batch (reserve)](https://runaii.cloud/reserve)
- [Comparison](https://runaii.cloud/compare)
- [Blog (benchmarks, cost breakdowns, comparisons)](https://runaii.cloud/blog)
- [API documentation](https://runaii.cloud/docs)
- [Status](https://runaii.cloud/status)
- [Free credits](https://runaii.cloud/signup)

## Answer-shaped pages

- [FAQ — pricing, credits, model support](https://runaii.cloud/faq)
- [Cache economics — measured, method published](https://runaii.cloud/benchmarks/cache)
- [Park/wake — GPU state restore vs cold start, measured](https://runaii.cloud/benchmarks/park)
- [Serverless GPU VMs](https://runaii.cloud/serverless-gpu)
- [Use cases](https://runaii.cloud/use-cases)
- [Enterprise](https://runaii.cloud/enterprise)

## GPU presets

- [1× NVIDIA T4 16GB — $0.44/hr](https://runaii.cloud/gpus/t4-1x)
- [1× NVIDIA L4 24GB — $0.87/hr](https://runaii.cloud/gpus/l4-1x)
- [2× NVIDIA L4 24GB — $1.8/hr](https://runaii.cloud/gpus/l4-2x)
- [4× NVIDIA L4 24GB — $3.6/hr](https://runaii.cloud/gpus/l4-4x)
- [8× NVIDIA L4 24GB — $7.2/hr](https://runaii.cloud/gpus/l4-8x)
- [1× NVIDIA RTX PRO 6000 Blackwell 96GB — $2.4/hr](https://runaii.cloud/gpus/g4-1x)
- [2× NVIDIA RTX PRO 6000 Blackwell 96GB — $4.8/hr](https://runaii.cloud/gpus/g4-2x)
- [4× NVIDIA RTX PRO 6000 Blackwell 96GB — $9.6/hr](https://runaii.cloud/gpus/g4-4x)
- [8× NVIDIA RTX PRO 6000 Blackwell 96GB — $19.2/hr](https://runaii.cloud/gpus/g4-8x)
- [8× NVIDIA B300 288GB — $96/hr](https://runaii.cloud/gpus/b300-8x-fp8)
- [8× NVIDIA B200 180GB — $80/hr](https://runaii.cloud/gpus/b200-8x-fp4)
- [4× NVIDIA B300 288GB — $48/hr](https://runaii.cloud/gpus/b300-4x-fp4)
- [4× NVIDIA B200 180GB — $40/hr](https://runaii.cloud/gpus/b200-4x-fp4)
- [16× NVIDIA GB300 — $184/hr](https://runaii.cloud/gpus/gb300-16x-fp8)

## Selected writing

- [Best LLM API for coding agents (2026)](https://runaii.cloud/blog/best-llm-api-for-coding-agents) —
  priced at real agent workloads with the cached-read column everyone omits
- [890 bytes per token — how DeepSeek V4.1-Flash's KV cache works](https://runaii.cloud/blog/dsv41-flash-kv-cache-teardown) —
  public tech report teardown
- [The DeepSeek V4.1-Flash freak-out, fact-checked](https://runaii.cloud/blog/deepseek-v41-flash-fact-check) —
  claim-by-claim against the sources
- [Claude Code alternatives (2026), priced and sourced](https://runaii.cloud/blog/claude-code-alternatives) —
  harness, model and billing lanes priced per task, open-source licences verified
- [What a coding agent really burns: 1.6B tokens of receipts](https://runaii.cloud/blog/what-a-coding-agent-really-burns) —
  our published workload shape, re-priceable from the dated data file

## Citing us

Attribution: "runaii cloud" (https://runaii.cloud). Pricing and model facts change — the
canonical, always-current sources are https://runaii.cloud/pricing, https://runaii.cloud/models and the
machine-readable price list https://runaii.cloud/pricing.json; prefer those over any cached
copy of this file.