A speed-dating app for free LLMs that actually benchmarks before you commit

What it does

free-coding-models is a CLI and web dashboard that catalogs ~153 free or free-limited coding LLMs from 16 providers (NVIDIA NIM, Groq, Cerebras, OpenRouter, Alibaba DashScope, etc.). It pings them all in parallel, surfaces live latency, and writes your chosen endpoint directly into tools like Aider, Goose, Continue, Cline, or OpenCode — so you skip the “create account, test, regret” cycle.

The interesting bit

The tool computes a Stability Score (0-100) that deliberately punishes flaky winners. It weights p95 latency, jitter, spike rate, and uptime — because a model that averages 800ms but randomly hits 6 seconds is worse than one that’s consistently 1.2s. The README also tracks its own audit trail: providers get removed when they go paid-only, die off, or prove “too tiny to be useful.”

Key highlights

• Live parallel benchmarking across 16 providers with tier rankings (S+ down to C) mapped to SWE-bench scores
• One-key setup: add a single NVIDIA_API_KEY or GROQ_API_KEY, skip the rest, expand later with P in the TUI
• Web dashboard with URL deep-linking (?tier=S+&sort=verdict&origin=groq) so filter states become shareable links
• Auto-writes endpoint configs for 15+ coding tools; the web UI never shells out to external binaries
• Docker image available, though GHCR requires authentication even for public pulls

Caveats

• The “free” label varies wildly: some providers need no credit card, others demand $10 spend for meaningful quotas, and GitHub Models ties to your Copilot tier
• The README’s provider table lists an iFlow shutdown date of “April 17, 2026” — either prescient or a typo; either way, the audit cleanup section admits the catalog is actively pruned
• Web dashboard claims “M4 parity with TUI” in one section and “M3 parity” in another; exact version alignment is unclear

Verdict

Worth it if you’re juggling multiple free-tier API keys and want data, not vibes, on which model to route to today. Skip if you already pay for one reliable provider and don’t need the comparison overhead.