Cheapest AI Models

The lowest-cost large language models you can call in production, ranked by input price. All available on one OpenAI-compatible API on thalam, pay per token, credits never expire.

  1. 1
    MT
    Llama 3.1 8B Instruct
    $0.020 / 1M · 128K context · Fast

    Meta's compact 8B Instruct, strong baseline for high-throughput backend chat and agent loops at near-zero cost.

  2. 2
    MI
    Mistral Nemo
    $0.040 / 1M · 128K context · Fast

    Cheapest model in the catalog. Mistral-Nemo for budget pipelines.

  3. 3
    AI
    GPT-OSS 120B
    $0.100 / 1M · 128K context · Fast

    OpenAI's open-weight 120B coding model. Apache-licensed, fully self-hostable.

  4. 4
    MT
    Llama 3.3 70B Instruct
    $0.130 / 1M · 128K context · Medium

    Meta's Llama 3.3 70B Instruct. Open-weight foundation, widely fine-tuned, reliable baseline.

  5. 5
    DS
    DeepSeek V4 Flash 0731
    $0.140 / 1M · 1M context · Fast

    The official DeepSeek V4 Flash build. Same size and price as the preview, substantially stronger on agentic and coding work.

  6. 6
    DS
    DeepSeek V4 Flash
    $0.140 / 1M · 1M context · Fast

    DeepSeek V4's fast variant. Same 1M context and reasoning mode, optimised for high concurrency and low latency.

Use any of these through one API

One OpenAI-compatible endpoint for every model above. Pay per token, no subscription, credits never expire.

FAQ

What is the cheapest LLM API?

Small open-weight models (Llama 3.1 8B, Mistral Nemo, GPT-OSS) cost a few cents per million tokens. On thalam you pay only for what you use with no platform fee or subscription, so the sticker price is the price.

Are cheap AI models good enough for production?

For classification, extraction, routing and high-volume simple tasks, small models are excellent and dramatically cheaper. A common pattern is a cheap default with a frontier model as fallback, on thalam that is one API and a config switch.

More best-of lists