Question 1

Which LLM API is cheapest in September 2026?

Accepted Answer

At list price, GPT-6 Luna ($0.10 input / $0.50 output per 1M tokens), Gemini 2.5 Flash-Lite ($0.10 / $0.40) and DeepSeek V4.1-Flash ($0.15 / $0.60 off-peak) are the cheapest hosted models. Among frontier models, GPT-6 Sol ($2 / $10) and Claude Opus 5.5 ($4 / $20) undercut GPT-6 Astra and Claude Fable 5.1 ($10 / $50).

Question 2

How is LLM API pricing calculated?

Accepted Answer

LLM APIs charge separately for input tokens (your prompt, system message, tools and retrieved context) and output tokens (the model's reply, including reasoning tokens), priced per 1 million tokens. Cached input tokens cost 5-10% of the input price on OpenAI and Anthropic, and batch APIs take a further 50% off.

Question 3

How often do LLM prices change?

Accepted Answer

Frequently. In September 2026 alone OpenAI halved prices with GPT-6 Sol and Luna, Anthropic cut Opus pricing by 40% with Opus 5.5, and Google set a promotional rate for Gemini 3.8 Flash that expires on 31 December 2026. This page records the date each figure was verified.

LLM API Pricing Comparison (September 2026)