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.
Combine asynchronous 50% Batch API pricing with medium reasoning output budgets.
Multi-turn history expands average input context to 6,990 tokens/req. Output accounts for 54% of total spend.
| Model | Delta | |||
|---|---|---|---|---|
| GPT-4.1 nano (Ultra-Low Cost)OpenAI | $0.10 | $0.40 | $47.04 | Save $488 |
| Claude Haiku 5.5 (High-Volume)Anthropic | $0.10 | $0.50 | $52.66 | Save $482 |
| DeepSeek V4.1 Flash (Off-Peak)DeepSeek | $0.15 | $0.60 | $68.36 | Save $467 |
| GPT-4o mini (Legacy Budget)OpenAI | $0.15 | $0.60 | $72.96 | Save $462 |
| GPT-4.1 mini (Balanced Fast)OpenAI | $0.40 | $1.60 | $188.17 | Save $347 |
| Llama 3.3 / 4 70B-Class (Hosted API)Open-Weight Host | $0.59 | $0.79 | $202.57 | Save $332 |
| Gemini 3.5 Flash-Lite (Budget Edge)Google | $0.30 | $2.50 | $223.72 | Save $311 |
| DeepSeek V4 Pro (Off-Peak Reasoning)DeepSeek | $0.66 | $1.98 | $257.96 | Save $277 |
| Gemini 3.8 Flash (Intro Paid Tier)Google | $0.75 | $3.75 | $394.94 | Save $140 |
| o3-mini (Reasoning)OpenAI | $1.10 | $4.40 | $535.01 | Active Baseline |
| Gemini 4 HydrogenGoogle | $1.50 | $7.50 | $785.1 | +$250 |
| GPT-5 (Flagship Reasoning)OpenAI | $1.25 | $10.00 | $904.78 | +$370 |
| GPT-4.1 (1M Long-Context)OpenAI | $2.00 | $8.00 | $940.84 | +$406 |
| Claude Sonnet 5.5 (Coding & Agents)Anthropic | $2.00 | $10.00 | $1,046.8 | +$512 |
| GPT-6.2 SolOpenAI | $2.00 | $10.00 | $1,046.8 | +$512 |
| Gemini 4 CarbonGoogle | $2.00 | $10.00 | $1,046.8 | +$512 |
| Gemini 3.1 Pro (Long-Context)Google | $2.00 | $12.00 | $1,184.67 | +$650 |
| GPT-4o (Legacy Omni)OpenAI | $2.50 | $10.00 | $1,215.94 | +$681 |
| Gemini 4 NeonGoogle | $3.00 | $15.00 | $1,570.2 | +$1,035 |
| Claude Opus 5.5 (Agentic Frontier)Anthropic | $4.00 | $20.00 | $2,093.59 | +$1,559 |
| Claude Fable 5.2Anthropic | $8.00 | $40.00 | $4,174.42 | +$3,639 |
| GPT-6.2 AstraOpenAI | $8.00 | $40.00 | $4,187.19 | +$3,652 |
Batch API cuts non-urgent nightly CI triage bills strictly in half. Pre-populated at 1,200 daily requests with 6,000 input tokens, 1,800 output tokens, 25% prompt caching, and 2 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.