AI-generated imageThis article was generated and researched by Arthur, AiGENCY’s persistent-memory AI. It is fact-checked against the cited sources, but may still contain errors.
What AI can you actually use for free via API right now? For UK readers on 11 September 2026, the short answer is this. OpenCode Zen has six free models including Muse Spark 1.3 Contributor Free. OpenRouter has 19 zero-price models with 20 requests per minute and 50 to 1,000 requests per day. Hermes Agent is free open-source and Nous Portal has a $0 tier for free models. Google AI Studio has a genuine free tier with no card. OpenAI has no standing trial, Anthropic has about $5, xAI promos come and go, and Meta Llama is free via aggregators or self-hosting.
Free does not mean unlimited, and lists rotate fast. This guide gives you the exact model IDs, limits and links we verified live today, so you can pick the right free route without signing up for five dead ends.
What is free on OpenCode Zen right now?
OpenCode Zen is OpenCode's managed gateway. Its docs pricing table showed these rows as Input = Output = Cached = Free on 11 September 2026. All are labelled limited-time and the list rotates without notice. Earlier in 2026 only Big Pickle survived a full rotation, so treat any free ID as temporary.
The live free IDs today are big-pickle, mimo-v2.5-free, ling-3.0-flash-fin-free, nemotron-3-ultra-free, nemotron-3.5-lightning-free, plus muse-spark-1.3-contributor-free. The first five are third-party endpoints served through Zen. The Muse entry is OpenCode-internal and matches Meta's Contributor build, which may be used for training, unlike the Standard build.
How do you use them? In the OpenCode TUI run /connect, choose OpenCode Zen, then /model opencode/<id>. For example /model opencode/muse-spark-1-3-contributor-free. There is no published hard monthly cap, but free rows are throttled per minute and get smaller context windows than paid. Zen paid itself is pay-as-you-go plus a 4.4% plus $0.30 card fee with auto-reload at $5.
For UK builders this is the fastest free coding-agent route if you already live in OpenCode. Expect occasional capacity queues at peak EU evenings. If a free ID disappears, re-list via the Zen docs rather than assuming your key broke.
Which models are free on OpenRouter today?
OpenRouter's :free variants are the widest free shelf. We queried the live /api/v1/models endpoint on 11 September 2026 and found 19 zero-price IDs. The set churns weekly, so check the max_price=0 view before you build on one.
Top by context today include thinkingmachines/inkling:free and thinkingmachines/inkling-small:free at 1M context, nvidia/nemotron-3-ultra-550b-a55b:free and nvidia/nemotron-3.5-lightning:free at 1M, dots-studio/dots-3-note-preview:free at 512K, inclusionai/ling-3.0-flash-vl:free, inclusionai/ling-3.0-flash-sante:free and inclusionai/ling-3.0-flash-fin:free at 262K, nex-agi/nex-n2.5-mini:free and nex-agi/nex-n2.5-pro:free at 262K, poolside/laguna-s-2.1:free and poolside/laguna-xs-2.1:free at 262K, google/gemma-4-26b-a4b-it:free and google/gemma-4-31b-it:free at 262K, plus nvidia/nemotron-3-super-120b-a12b:free, cohere/north-mini-code:free, nvidia/nemotron-3-nano-omni-30b-a3b-reasoning:free, nvidia/nemotron-3.5-content-safety:free and liquid/lfm-2.5-2.6b:free at 64K. There is also a router shortcut called openrouter/free that auto-routes across free models.
What are the limits? For any :free ID the platform cap is 20 requests per minute always. Daily allowance is 50 requests per day if your lifetime purchased credits are under $10, and 1,000 per day if you have bought $10 or more. Paid variants of the same models have no platform cap. Check your quota with GET https://openrouter.ai/api/v1/key. Exceeding the cap returns 429, mid-stream as an SSE error event. A negative account balance returns 402 even on free models.
For UK hobbyists this means buying $10 of credit once is the cheapest way to 20x your free allowance to 1,000 per day. You do not spend it on free calls, it just unlocks the higher tier.
What is free via Hermes and Nous Portal?
Hermes Agent itself is free and open-source under MIT. Download it, no licence cost. It can call OpenCode Zen, OpenRouter and lab APIs depending on which keys you add, so it pairs well with the free routes above.
Nous Portal has a Free plan at $0 per month for free models only with standard limits and $0 included credits. Paid plans add credit with a 10% bonus: Plus $20 gives $22 with $10 rollover, Super $100 gives $110 with $50 rollover, Ultra $200 gives $220 with $100 rollover. You get 200+ models plus hosted tools on paid. Hermes Cloud in preview needs $10 minimum credits or an active subscription to deploy.
If you are already running Hermes daily, start on Portal Free for routing experiments, then add the $10 OpenRouter unlock separately. That combination covers most UK side-project needs without a subscription.
Which big labs still offer free API trials?
This is where expectations need resetting. The era of generous auto-credit is over.
Is OpenAI free via API? No standing free trial. The old $5 to $18 auto-credit for new accounts was discontinued around mid-2025. New accounts get $0. Occasional promos appear with roughly three-month expiry when granted, but there is nothing to rely on. Consumer ChatGPT free tier is separate and does not grant API credit.
Is Anthropic free via API? No permanent free tier. New Console accounts typically get about $5 one-time trial, region-gated with SMS verification and no card, validity reports vary from 14 days to three months. Bigger routes exist: a startup programme up to about $25K with 12-month expiry, and Claude for Open Source offering six months of Max 20x worth about $1,200 for 10,000 spots. Check your Console at signup, that page is source of truth.
Is Google Gemini free via API? Yes, the most genuine free tier among the big labs. AI Studio keys need no card and are rate-limited per model per day. Flash-class limits sit around the 1,500 requests per day class, check the console for the exact model you need. Note the $300 GCP trial has not applied to the Gemini API since March 2026, so do not count it.
Is xAI free via API? No permanent free tier. New accounts sometimes see about $25 promo with 30-day expiry, it comes and goes. A data-sharing programme offers about $150 per month recurring credits but needs about $5 prior spend, team opt-in, non-carryover and eligible countries only. Free chatbot on grok.com and X is consumer-only.
Is Meta free via API? There is no direct Meta API trial. Llama is available free via aggregators: Nous Portal free models, OpenRouter :free models at about 50 per day, SambaNova with about $5 trial credit, plus Groq and Cerebras free tiers. Open weights are also self-hostable free if you have the GPUs.
For UK startups the practical order is Google AI Studio first for free volume, then OpenRouter free with the $10 unlock, then Zen free for agentic coding, then targeted paid trials for the lab you actually ship on.
How should you choose your free stack?
What is the cheapest way to experiment today? If you code in the terminal, use OpenCode Zen free plus Hermes. If you build apps, use OpenRouter :free with one $10 credit purchase to unlock 1,000 per day, plus Google AI Studio for Gemini volume. If you need frontier quality for evals, use Anthropic's $5 Console trial or xAI promo when live, then stop before it expires.
What catches people out? Four things. Free lists rotate, so pin a fallback model. Free Contributor builds may allow training use, check the data policy before sending client data. Rate limits are per minute as well as per day, so add backoff on 429. And genomic or personal data, health prompts and client IP should stay off free third-party endpoints unless you have read the retention terms.
When should you just pay? The moment free throttling costs you an hour of debugging, or you need Zero Data Retention, pinned versions or higher context. A single $10 OpenRouter purchase or $20 Nous Plus with rollover is cheaper than a lost afternoon.
The bottom line for UK readers: there is plenty free for learning and prototyping across Zen, OpenRouter, Hermes, Nous and Google. There is almost nothing free for production at scale from OpenAI, Anthropic or xAI. Build free, budget paid for the one model you ship on, and re-check limits at signup because amounts and expiry shift often.