Mesh Research · 4 published

    Mesh API 
    Research Labs.

    The research arm of Mesh API. Reproducible benchmarks, cost autopsies, and routing studies, built to be cited, criticised, and rerun.

    Get your API key →
    Save 95% Claude Sonnet 4.5 Tokens
    Featured studyJun 12, 202614 min read
    Engineering Notes

    Save 95% Claude Sonnet 4.5 Tokens

    Engineering notes from building a beta token-savings dial for the Mesh gateway. Prompt caching mechanics, an honest savings receipt, real terminal output, and the 500 error that taught us our own schema.

    1 dial 0–0.95~95% input savings ceilingBetaMIT reference impl
    Raushan SharmaRead study
    Mesh API Value Ranking: capability per dollar.
    BenchmarkJun 5, 202612 min read
    Value Rankings

    Mesh API Value Ranking: capability per dollar.

    The full price-performance leaderboard for every text model on Mesh. 122 ranked on verified Artificial Analysis indices, 176 more listed honestly on a data-pending shelf with real Mesh pricing.

    1,000+ models122 ranked64x price spreadTop value: DeepSeek V4 Flash
    Raushan SharmaRead study
    You can pay 7.5x more for a worse image.
    BenchmarkMay 24, 20268 min read
    Dollar per Image Bench

    You can pay 7.5x more for a worse image.

    A cost x quality map of 7 frontier image models on Mesh API. The cheapest image beats the most expensive at 1/7.5th the price.

    7 models7.5x cost spread9.1x latency spread$1.27 spent
    Raushan SharmaRead study
    Same answer. Up to 1000× the bill.
    BenchmarkMay 24, 202610 min read
    Dollar per Task Bench

    Same answer. Up to 1000× the bill.

    Five LLMs, ten real tasks, one API. The cheapest model passed every coding test and matched on support quality, while the bill swung 259× across the same session.

    5 models · 10 tasks259× session gap96% unseen tokens$0.43 spent
    Raushan SharmaRead study

    Asia's largestsovereign AI gateway.

    1,000+ AI models. One endpoint. Your data stays inside your border.

    Get an API key