BenchLM selector matches models to tasks, budgets and deployment needs

BenchLM’s LLM selector recommends candidates from 242 models based on the work users describe, their budget, input size and deployment requirements. It covers 32 tasks for everyday users and builders, using relevant benchmark evidence to rank candidates that meet those constraints. Users can describe a job in plain language or choose options directly.

A decision model translates descriptions into editable form answers but never sees scores or chooses recommendations. The shortlist is computed on the page from a public dataset, with assumed defaults clearly marked. Results remain provisional while assumptions or setup requirements need confirmation. Unsupported tasks, including some voice, image generation and specialist research work, receive an evidence-gap notice rather than recommendations based on unrelated capabilities.

Budget comparisons use a stated workload of 1M input + 200k output tokens, excluding caching, retries, tools and special tiers. Local-deployment requirements restrict candidates to open-weight models, but the selector does not verify hardware compatibility. API prices also do not establish subscription access, message limits or app features.

The selector is free and requires no account, although anonymous description readings have a daily limit. BenchLM retains descriptions, including drafts, for 180 days without an IP address or account. Shared links contain selected answers rather than descriptions and rerun those answers against current data.