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.
Visualize triangular input token growth across long un-pruned customer conversations.
Multi-turn history expands average input context to 6,060 tokens/req. Output accounts for 27% of total spend.
| Model | Delta | |||
|---|---|---|---|---|
| GPT-4.1 nano (Ultra-Low Cost)OpenAI | $0.10 | $0.40 | $39.18 | Save $982 |
| Claude Haiku 5.5 (High-Volume)Anthropic | $0.10 | $0.50 | $40.92 | Save $980 |
| DeepSeek V4.1 Flash (Off-Peak)DeepSeek | $0.15 | $0.60 | $56.48 | Save $964 |
| GPT-4o mini (Legacy Budget)OpenAI | $0.15 | $0.60 | $61.26 | Save $960 |
| Gemini 3.5 Flash-Lite (Budget Edge)Google | $0.30 | $2.50 | $150.16 | Save $871 |
| GPT-4.1 mini (Balanced Fast)OpenAI | $0.40 | $1.60 | $156.71 | Save $864 |
| Llama 3.3 / 4 70B-Class (Hosted API)Open-Weight Host | $0.59 | $0.79 | $217.53 | Save $803 |
| DeepSeek V4 Pro (Off-Peak Reasoning)DeepSeek | $0.66 | $1.98 | $231 | Save $790 |
| Gemini 3.8 Flash (Intro Paid Tier)Google | $0.75 | $3.75 | $306.91 | Save $714 |
| o3-mini (Reasoning)OpenAI | $1.10 | $4.40 | $449.22 | Save $572 |
| Gemini 4 HydrogenGoogle | $1.50 | $7.50 | $608.85 | Save $412 |
| GPT-5 (Flagship Reasoning)OpenAI | $1.25 | $10.00 | $614.25 | Save $407 |
| GPT-4.1 (1M Long-Context)OpenAI | $2.00 | $8.00 | $783.57 | Save $237 |
| Claude Sonnet 5.5 (Coding & Agents)Anthropic | $2.00 | $10.00 | $811.8 | Save $209 |
| GPT-6.2 SolOpenAI | $2.00 | $10.00 | $811.8 | Save $209 |
| Gemini 4 CarbonGoogle | $2.00 | $10.00 | $811.8 | Save $209 |
| Gemini 3.1 Pro (Long-Context)Google | $2.00 | $12.00 | $873.23 | Save $148 |
| GPT-4o (Legacy Omni)OpenAI | $2.50 | $10.00 | $1,020.97 | Active Baseline |
| Gemini 4 NeonGoogle | $3.00 | $15.00 | $1,217.7 | +$197 |
| Claude Opus 5.5 (Agentic Frontier)Anthropic | $4.00 | $20.00 | $1,623.59 | +$603 |
| Claude Fable 5.2Anthropic | $8.00 | $40.00 | $3,233.91 | +$2,213 |
| GPT-6.2 AstraOpenAI | $8.00 | $40.00 | $3,247.19 | +$2,226 |
Sliding-window context trimming and prompt caching prevent runaway multi-turn bills. Pre-populated at 1,800 daily requests with 3,000 input tokens, 500 output tokens, 20% prompt caching, and 10 turns/session.
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.