Cost calculations for LLM providers

llm-pricingapi-costsmodel-comparison

Abstraction: Comparative USD-per-million-token pricing table with LMSYS ELO scores

Key points:

All prices in USD per 1M tokens; table shows input, output, and three input/output balance scenarios (50/50, 25/75, 75/25)
Cheapest options: Anyscale Mistral-7B at $0.15/1M flat; GPT-4o-mini at $0.15 input / $0.60 output (~$0.38 balanced)
Mid-tier: GPT-3.5 at $0.50/$1.50, Gemini Pro at $1.00/$2.00, Claude Instant at $0.80/$2.40 per 1M tokens
Premium tier: GPT-4-turbo at $10/$30, Claude 2.0/1 at $8/$24, GPT-4o at $5/$15 per 1M tokens
LMSYS ELO scores included: GPT-4o (1310), GPT-4-turbo (1249), Mistral medium (1150), GPT-3.5 (1116); GPT-4o offers the best ELO-per-dollar ratio among frontier models
Example prompt of 300 tokens input / 100 tokens output used to illustrate per-call costs; GPT-4-turbo example cost $0.006 vs GPT-4o-mini $0.000105

Connections: Openai · Anthropic · Google · Mistral · Large Language Models · LLM Pricing