Cloud vs. Rig TCO
Plot cumulative 36-month API spending against hourly RunPod/H100 rentals and owned RTX/Mac hardware with electricity and resale value.
Model real-world token billing including prompt caching discounts, 50% async Batch API tiers, multi-turn conversation replay bloat, and hybrid frontier-to-flash model routing.
Multi-turn history expands average input context to 5,880 tokens/req. Output accounts for 58% of total spend.
| Model | Delta | |||
|---|---|---|---|---|
| GPT-4.1 nano (Ultra-Low Cost)OpenAI | $0.10 | $0.40 | $45.91 | Save $872 |
| Claude Haiku 5.5 (High-Volume)Anthropic | $0.10 | $0.50 | $47.21 | Save $870 |
| DeepSeek V4.1 Flash (Off-Peak)DeepSeek | $0.15 | $0.60 | $59.61 | Save $858 |
| GPT-4o mini (Legacy Budget)OpenAI | $0.15 | $0.60 | $78.94 | Save $839 |
| GPT-4.1 mini (Balanced Fast)OpenAI | $0.40 | $1.60 | $183.66 | Save $734 |
| Gemini 3.5 Flash-Lite (Budget Edge)Google | $0.30 | $2.50 | $194.91 | Save $723 |
| DeepSeek V4 Pro (Off-Peak Reasoning)DeepSeek | $0.66 | $1.98 | $229.49 | Save $688 |
| Llama 3.3 / 4 70B-Class (Hosted API)Open-Weight Host | $0.59 | $0.79 | $306.06 | Save $611 |
| Gemini 3.8 Flash (Intro Paid Tier)Google | $0.75 | $3.75 | $354.11 | Save $563 |
| o3-mini (Reasoning)OpenAI | $1.10 | $4.40 | $578.89 | Save $339 |
| Gemini 4 HydrogenGoogle | $1.50 | $7.50 | $688.09 | Save $229 |
| GPT-5 (Flagship Reasoning)OpenAI | $1.25 | $10.00 | $789.93 | Save $128 |
| Claude Sonnet 5.5 (Coding & Agents)Anthropic | $2.00 | $10.00 | $917.45 | Active Baseline |
| GPT-6.2 SolOpenAI | $2.00 | $10.00 | $917.45 | +$0 |
| Gemini 4 CarbonGoogle | $2.00 | $10.00 | $917.45 | +$0 |
| GPT-4.1 (1M Long-Context)OpenAI | $2.00 | $8.00 | $918.3 | +$1 |
| Gemini 3.1 Pro (Long-Context)Google | $2.00 | $12.00 | $1,050.82 | +$133 |
| GPT-4o (Legacy Omni)OpenAI | $2.50 | $10.00 | $1,315.66 | +$398 |
| Gemini 4 NeonGoogle | $3.00 | $15.00 | $1,376.17 | +$459 |
| Claude Opus 5.5 (Agentic Frontier)Anthropic | $4.00 | $20.00 | $1,834.89 | +$917 |
| Claude Fable 5.2Anthropic | $8.00 | $40.00 | $3,616.1 | +$2,699 |
| GPT-6.2 AstraOpenAI | $8.00 | $40.00 | $3,669.79 | +$2,752 |
Baseline uncached rates vs. cached prompt hit discounts across major providers. Output tokens carry 3× to 8× higher cost per token than inputs, making structured output conciseness the fastest optimization lever.
| Model | Input / 1M | Output / 1M | Cached Discount | Context Window |
|---|---|---|---|---|
| GPT-5 (Flagship Reasoning) | $1.25 | $10.00 | 90% off input | 400K |
| Claude Sonnet 5.5 (Coding & Agents) | $2.00 | $10.00 | 95% off input | 1,000K |
| Claude Opus 5.5 (Agentic Frontier) | $4.00 | $20.00 | 95% off input | 1,000K |
| Gemini 3.8 Flash (Intro Paid Tier) | $0.75 | $3.75 | 90% off input | 1,000K |
| Gemini 3.1 Pro (Long-Context) | $2.00 | $12.00 | 90% off input | 1,049K |
| DeepSeek V4.1 Flash (Off-Peak) | $0.15 | $0.60 | 98% off input | 1,000K |
Daily cost = ((Daily_Input_Tokens / 1M) * Effective_Input_Rate) + ((Daily_Output_Tokens / 1M) * Output_Rate), scaled over 30.4375 average calendar days per month and adjusted for multi-turn chat replay growth.
Primary Technical & US Grid Benchmarks
Plot cumulative 36-month API spending against hourly RunPod/H100 rentals and owned RTX/Mac hardware with electricity and resale value.
Count live prompt tokens across English prose and JSON/code, size RAG document chunks, and compute float32 vs. int8 HNSW vector RAM.
Test Llama 70B, Qwen 32B, or DeepSeek R1 across RTX 3090/4090/5090 & Apple Silicon Macs with GQA KV-cache math and tokens/sec speed.