LLM cost guides by use case

Each guide fixes one realistic workload, states its assumptions, and ranks every model by cost using the same engine as the calculator. Prices verified 2026-10-18.

• Cheapest LLM for coding agents

  What a 30-step coding agent run costs on every major model with prompt caching, and why step count matters more than the per-token price. Cheapest today: GPT-6 Luna (Per run (cached): $0.05).

• Cheapest LLM for RAG chatbots

  Monthly cost of a retrieval-augmented chatbot on every major model, with the cached system prompt and the effect of retrieving more context. Cheapest today: Claude Haiku 5.5 (Per 1,000 answers: $0.67).

• Cheapest LLM for batch classification and extraction

  Cost to classify 1M documents or extract JSON from 100K documents on every major model, with batch API discounts and DeepSeek off-peak pricing. Cheapest today: Claude Haiku 5.5 (Classify 1M docs: $35.00).

• Long-context document analysis (400K tokens)

  What it costs to ask questions about a 400K-token document on every model that fits it, including long-context surcharges and prompt caching. Cheapest today: DeepSeek V4.1 Flash (10 questions, cached: $0.17).

• Cheapest LLM for high-volume customer support

  Monthly cost of one million support-bot replies on every major model, with a cached help-centre prompt and a small-model-plus-escalation setup. Cheapest today: Ministral 3 8B (Per 1,000 replies: $0.22).

• Prompt caching explained, with savings

  How prompt caching is billed by each provider, when a cache-write premium pays off, and how much caching saves on agents and chatbots.

The rankings compare price, not quality. Use them to build a shortlist, then test the shortlisted models on your own data.