即時語音與世界模型 API 成本指南:Genie 4 與 GPT-Live-1.1 待機省費實測
深度測算串流多模態 API 帳單:Google Genie 4 720p24fps 每秒 $0.10(約 $6.05/分鐘),OpenAI GPT-Live-1.1 WebRTC 待機模式直降 60% 至 $0.02/分鐘。
OpenAI GPT-Live-1.1 Standby Calculator
Active $0.05/min ($3.00/hr) vs WebRTC Standby $0.02/min (-60%)
Google Genie 4 World Model Simulator
720p · 24fps · 5,760 video tok/s ($17.50/M out, $1.50/M in)
Tip: Video frames dominate >99% of Genie 4 billing—cap stream TTL and cascade non-visual reasoning to text routers.
Real-Time Multimodal Unit Economics Comparison
Verified Live 2026 Rates| Model / Mode | Billing Unit | Raw Rate | Per Minute Eq. | 10,000-Min Est. | Best Use Case |
|---|---|---|---|---|---|
| GPT-Live-1.1 (Active) | Per-Second Session | $0.05 / min | $0.0500 / min | $500.00 | Full-duplex speech & live turn-taking |
| GPT-Live-1.1 (WebRTC Standby -60%) | Idle Connection (5–60s) | $0.02 / min | $0.0200 / min | $200.00 | Persistent voice channels & pause-heavy agents |
| Google Genie 4 (720p 24fps) | 5,760 video tok/s | $17.50 / 1M out | ~$6.05 / min | $60,500.00 | Interactive 3D world simulation & gaming QA |
| Groq LPU + Whisper Pipeline (OSS) | LPU Token + Audio | $0.59 / $0.79 (1M) | ~$0.004 / min | ~$40.00 | 420 t/s sub-200ms TTFT modular voice stack |
Next Step: Compare Orchestration Flagships & Cascade Costs
Compare Gemini 4 Flagship Tiers
Gemini 4 Neon ($3/$15, 61.4 AA) vs GPT-6.2 Sol ($2/$10) orchestration cost.
Audit OpenAI Flagship Stack
Benchmark GPT-6.2 Astra vs Claude Opus 5.5 reasoning & cache savings.
Simulate Text Cascade Cost
Offload background RAG & tool routing to DeepSeek-V4.1-Flash ($0.006/M cache).
Frequently Asked Questions
1How is Google Genie 4 API pricing calculated per second and per minute?
Google Genie 4 (models/genie-4 on the Gemini v1beta Interactions API) bills text/reference image input at $1.50/1M tokens, text output at $9.00/1M tokens, and synchronized 720p/24fps video + 48kHz stereo audio at $17.50/1M tokens. Because each 720p frame consumes 240 tokens (24 fps × 240 = 5,760 tokens/sec), live streaming video output costs roughly $0.10 per second (~$6.05 per minute) on Paid Tier—Free, Batch, Flex, and Priority tiers are not supported.
2How much does OpenAI GPT-Live-1.1 WebRTC Standby mode save on voice agents?
Active sessions on gpt-live-1.1 (/v1/live/sessions, up to 2 hours per session) cost $0.05/min ($3.00/hr). Enabling WebRTC Standby with a 5–60 second inactivity timeout drops idle connection billing by 60% to $0.02/min ($1.20/hr), billed per second across both gpt-live-1 and gpt-live-1.1.
3When should teams cascade to text/tool routers instead of keeping Genie 4 or GPT-Live active?
Offload background RAG retrieval, state summarization, and classification to sub-cent text models like deepseek-v4.1-flash ($0.0498/$0.1992 per 1M on OpenRouter, $0.001/M cached, with built-in web search/fetch tool fees removed) or gpt-oss-120b ($0.09/$0.36 per 1M), keeping real-time voice or world-model streams reserved strictly for active user interaction windows.
4Can Prompt Caching or Batch 50% discounts offset Genie 4 streaming bills?
Genie 4 explicitly excludes Batch, Flex, and Free tiers, meaning per-frame video output cannot be discounted by 50% Batch queues. Teams lower overall unit economics by caching static scene prompts/system rules on text orchestration models (gemini-4-neon at $0.15/1M cached) and enforcing strict per-session duration caps.
12 月 1 日漲價前需要企業級大模型遷移與快取省錢審計?
正在大規模使用 Gemini 4 Carbon、受 Azure o3/o4-mini 停服影響,或面臨即時語音/世界模型串流帳單壓力?立即預約 TokenCosts AI 架構成本審計,或直接發郵件至 [email protected] 聯繫工程團隊。