Is the Kimi K2 API free? Moonshot's endpoint is paid, but the open-weights model runs at $0 via OpenRouter, Cloudflare Workers AI, and Claude Code.

SambaNova Cloud is the free fast-inference API for people who need big models to move quickly. Where Groq and Cerebras win on small-model speed, SambaNova runs…

We measured Zhipu GLM 4 Flash at 20 tok/s, the slowest free AI API we tested. Here is where it still makes sense: Chinese text, no credit card, Claude Code backend.

LiteLLM is a free, open-source gateway to call OpenAI, Claude, Gemini and 100+ LLMs through one API. SDK, self-hosted proxy, fallbacks, budgets.

Firecrawl turns any URL into clean, LLM-ready Markdown or JSON via one API. Free tier: 1,000 credits/month, no card. Scrape, crawl, map & extract.

The Grok API has no permanent free tier, but xAI data sharing gives up to $150/month in free credits. How it works, pricing, and OpenAI-SDK setup.

The exact MCP tool names for all three: brave_web_search and seven more, tavily-search with hyphens, web_search_exa with the provider last. Plus what each free tier gives you.

We measured Groq free TTS at 4.4x realtime with 6 voices and no card, plus two traps: the model needs manual terms acceptance, and its WAV header reports 24 hours for a 10-second clip.

We timed Groq free Whisper on real speech: 4 minutes of audio transcribed in 1.53 seconds, 160x realtime. Turbo was 21% faster than large-v3 with identical output.

Together AI new accounts are read-only until you deposit, and the -Free model suffix no longer exists. What we hit when benchmarking, the current per-token pricing, and the card-free alternatives.