Free tier

AI Gateway · the documented free allowances behind auto/free — verified numbers, live headroom, and the always-on cascade for long-horizon tasks.

documented free tokens per month · wired providers
≈ 46.0M
Groq chat models 1.4M documented tokens/day ×30 + Cloudflare ≈4.0M/month (10k neurons/day on llama-3.1-8b) — NVIDIA one-time credits excluded. Our arithmetic, from provider docs — sources on every card. Updated 2026-07-23. For scale: OmniRoute claims “💰 ~1.53B Free Tokens / Month” across its 90+ free providers — that's their number, across far more providers than the 3 wired here.
The always-on cascade — what model "auto/free" actually does
Your IDE / CLI / agent Cursor · Cline · Codex · any OpenAI SDK <origin>/v1 · model: auto/free LLM Switchboard Gateway router-ranked · cooldowns · headroom TIER 1 · HEADROOM fullest free pool Groq free tier 429? cooldown → TIER 2 · NEXT POOL other free allowances Cloudflare · NVIDIA quota out? → TIER 3 · RANKED REST anything else you enabled full catalog ● ALWAYS ON auto-fallback cascade · rate-limited pools cool down and recover on their own — the task keeps running
Free pools — documented allowance vs today's gateway usage — bars fill as this gateway meters real traffic (streams estimated)
groq
1.4M tok/day
Free Plan: per-model daily token caps on 6 chat models (llama-3.1-8b 500K · llama-3.3-70b 100K · gpt-oss-120b/20b/safeguard 200K each · qwen3.6-27b 200K)
today's gateway usage: loading…
per-model TPD + 30 RPM / 6-12K TPM caps; kimi-k2 and compound have no free token cap (compound: 250 req/day); spread work across models to use the full pool · source ↗
cloudflare
133k tok/day
Workers AI: 10,000 neurons/day free on every plan — ≈133K output tokens/day on llama-3.1-8b (≈49K/day on llama-3.3-70b)
today's gateway usage: loading…
neuron-denominated: token yield varies by model; exceeding stops or bills at $0.011/1K neurons on paid · source ↗
nvidia
non-token allowance
build.nvidia.com: one-time 1,000 API credits on sign-up (up to 5,000 total), 1 credit = 1 request — NOT a renewing monthly allowance
today's gateway usage: loading…
one-time trial credits, excluded from monthly totals; great for burst evaluation, not steady long-horizon load · source ↗
Run a long-horizon task on the free pools
1. Point your agent at https://llm-switchboard.icompaas.com/v1 with your provider key as the Bearer token.
2. Set model: "auto/free" — candidates come only from free-tier providers, ordered by headroom (the pool with the most remaining documented allowance goes first).
3. When a pool rate-limits (429), it cools down honoring Retry-After and the next pool answers — the task keeps running. Watch X-Switchboard-Model / -Fallbacks to see it happen.
4. Scope which pools participate with the toggles on Connections (sent as X-Switchboard-Providers).
curlcurl https://llm-switchboard.icompaas.com/v1/chat/completions -H "Authorization: Bearer $GROQ_API_KEY" -H "Content-Type: application/json" -d '{"model": "auto/free", "messages": [{"role": "user", "content": "…long-horizon step…"}]}'
More documented free tiers — not wired here (yet), verified from provider docs
ProviderWhat's freeDailyMonthly est.Caveats · source
OpenRouter :free model variants (model IDs ending in :free) across the catalog, request-capped per day 50 req/day 20 RPM; 50 req/day if <$10 credits ever purchased, 1000 req/day with >=$10 purchased; API
Google AI Studio / Gemini API Free tier: 'Free of charge' input & output on 8 Flash-family models — Gemini 3.6 Flash, 3.5 Flash, 3.5 Flash-Lite, 3.1 Flash-Lite, 2.5 Flash, 2.5 Flash-Lite, 2.0 Flash, 2.0 Flash-Lite; embedding input also free (Gemini Embedding / Embedding 2) API key required; per-model RPM/TPM/RPD no longer published in docs (rate-limits page last
Cerebras Free tier inference on 3 models (gpt-oss-120b, zai-glm-4.7, gemma-4-31b): 1M tokens/day per model 1.0M/day 30.0M/mo Per model: 5 RPM, 30K TPM, 1M tokens/hour, 1M tokens/day; API key required; separate one-t
Mistral La Plateforme Free mode (default account state): rate-limited access to the API for testing/prototyping — no public token or request numbers API key required; Mistral publishes no free-tier numbers — help article says to view actua
Hugging Face Inference Providers $0.10/month included credits for free accounts ($2.00/month for PRO; $2.00/seat for Team/Enterprise) usable across 200+ models via HF-routed requests HF token required; credits are dollar-denominated ('$0.10, subject to change'), monthly re
Pollinations Keyless anonymous access to text/image/audio generation endpoints (1 request per 15s, basic models); free Seed registration raises to 1 request per 5s No signup required for Anonymous tier; interval-based limits (per 15s/5s), no token or dai
SambaNova Cloud Free Tier (no payment method linked): 200K tokens/day per model on DeepSeek-V3.1, Meta-Llama-3.3-70B-Instruct, gpt-oss-120b (+ preview DeepSeek-V3.2, gemma-4-31B-it) 200k/day 6.0M/mo API key required; hard cap of 20 RPM AND only 20 requests/day per model — hitting the 200K
Honesty box: every number here is documented-figures arithmetic (no extrapolation); one-time credit grants are excluded from monthly totals; the usage meter counts exact tokens for non-streamed calls and estimates streams (~4 chars/token); provider terms change — each card links its source.
Sign in to continue

LLM Switchboard is private — sign in with Authlee to access the control room.

Sign in with Authlee
← Back to home