anytimelime — the free AI field report

What's free in AI right now — 2026-09-03

Eight models cleared the tool‑call test today, giving you a tiny but solid baseline. The coder‑flavored cohere/north-mini-code:free (4.7s) is the only entry that actually typed a function, while the ultra‑fast liquid/lfm-2.5-2.6b:free (0.8s) shows latency can be under a second. The MiniMax pair – minimax/minimax-m2.7:free (1.9s) and minimax/minimax-m3:free (2.3s) – both passed, as did the Nvidia‑hosted deepseek-ai/deepseek-v4-pro-0813 (2.6s). The rest are gated: SambaNova’s payment wall, Nvidia’s empty responses, or we burned our rate limit on OpenRouter’s untested freebies.

Failure modes stay boring: SambaNova still shouts “add a payment method,” Nvidia returns nothing, and some OpenRouter models sit untested. Today’s eight give you a quick pipeline, but treat them like a moving lattice – the roster flips weekly, and the next probe could be a different eight or even fewer passes. Keep your scripts flexible; a missed tool call can break a flow faster than a stale dependency.

How to use them with AnytimeLime

Clone the engine from github.com/fusionpoweredsoftware/anytimelime, set LIME_KEY to your OpenRouter key, and run ./lime.sh. It pulls today’s verified roster and rotates across the passing models so no single free tier carries all your traffic. Your key lives on your machine – the website never asks for it.

If you already have an OpenRouter key, the model IDs above (e.g., cohere/north-mini-code:free, dots-studio/dots-3-note-preview:free, google/gemma-4-31b-it:free, inclusionai/ling-3.0-flash-fin:free, liquid/lfm-2.5-2.6b:free, minimax/minimax-m2.7:free, minimax/minimax-m3:free, deepseek-ai/deepseek-v4-pro-0813) work directly with any OpenAI‑compatible client. No AnytimeLime needed if you prefer to wire it up yourself.

This week's deep dig

Our speculative hunt for keyless gateways (BazaarLink, LLM7, community proxies) didn’t surface in today’s measured set. The eight PASS entries remain the only concrete evidence we have; the rest of the “free AI” ecosystem is still out of our probe’s reach. The old failure patterns hold, so today’s data validates the thesis that the measured table is the sole source of truth.

The churn

Free rosters change weekly – models get gated, die, or resurrect without notice. Today’s eight are tomorrow’s blanks. Re‑run the probe before you ship a pipeline that leans on a free model; a single missed tool call can break your flow faster than a stale cache.

Today's passing roster

• liquid/lfm-2.5-2.6b:free — openrouter · 0.8s · agentic 3/5
• google/gemma-4-31b-it:free — openrouter · 1.3s · agentic 3/5
• dots-studio/dots-3-note-preview:free — openrouter · 1.4s · agentic 3/5
• inclusionai/ling-3.0-flash-fin:free — openrouter · 1.5s · agentic 3/5
• minimax/minimax-m2.7:free — openrouter · 1.9s · agentic 3/5
• minimax/minimax-m3:free — openrouter · 2.3s · agentic 3/5
• deepseek-ai/deepseek-v4-pro-0813 — nvidia · 2.6s · agentic 3/5
• cohere/north-mini-code:free — openrouter · 4.7s · agentic 3/5

The measured table

Click any model for its endpoint and ready-to-run snippets in curl, Python, Go, Rust, and more.

8 made a tool call, 0 answered in text only, of 24 discovered. Numbers are single-run measurements from one location — your mileage will vary. Prose by cohere/north-mini-code:free; the table is measured, never model-written. Reproduce with ./free-scan.sh from the engine repo.