Find the best AI API routers.
Use them all with one API key.

Openference API offers a high-performance inference platform with a 3-day free trial for agentic development. It features subscription tiers (Lite/Pro/Max) with bundled requests and on-demand rates for extra usage.

Free plan: 3-day trial with open-source models — no credit card required

SambaNova Cloud provides fast LLM inference on specialized RDU hardware. It offers an OpenAI-compatible API with a persistent free tier (Developer Tier) and pay-as-you-go credits for higher rate limits.

One-time $5 signup credit for testing; persistent free tier limited by rate, not total dollar value.

Vercel AI Gateway is a unified API gateway that allows routing to hundreds of models from various providers through a single OpenAI-compatible endpoint. It provides observability, budgets, and zero-data-retention (ZDR) options with no token markup over upstream list prices.

Each team account gets $5 of AI Gateway credit every 30 days to try any model; after your first payment you become a paid customer and stop receiving the free credit. Official figure.

Dahl Inference is a high-performance platform offering decentralized access to open models like DeepSeek and MiniMax. It features an OpenAI-compatible API and a generous signup promotion of 100 million free tokens.

Value based on 100M free tokens at an estimated rate of $0.03/1M.

Qiniu Cloud AI LLM Inference is a managed MaaS platform that provides a unified, compliant API for over 50 mainstream LLMs. It is compatible with both OpenAI and Anthropic protocols, offering new users substantial free tokens and flexible resource packages for production scaling.

First API key activates a shared 3M-token free resource pack (~$3 at $1/1M; auto-granted ~1h after first billable usage), then PAYG; several open-source models are permanently fully free (bonus, unquantified).

SiliconFlow (SiliconCloud) is a premier Chinese AI model platform offering high-performance inference for a wide range of open-source models (Qwen, Llama, DeepSeek). It is notable for providing free, permanent API access to models with parameters under 9B, alongside competitive pay-as-you-go rates for larger flagship

Signup grants ~CNY 14 (≈$2.08) credit, plus ~CNY 16 (≈$2.38) voucher after real-name verification; many <=9B open-source models (Qwen2.5-7B, Llama3.1-8B, etc.) are permanently free; direct access within mainland China. Since 2026-05-15, unverified users are capped at 10 RPM and delinquent/unverified accounts are cut off.

The official Moonshot AI developer platform for Kimi models. Offers pay-as-you-go access to Kimi K3, K2.7-Code, and K2.6 over an OpenAI-compatible API. One-off trial credits available upon verification.

Not an ongoing free tier. New users who register and complete personal real-name verification receive roughly CNY 15 (≈$2.23) in one-off trial credits (time-limited), usable on the K2 family. The flagship K3 requires a prior top-up to unlock (community reports put the floor around CNY 10 (≈$1.49)), and the CNY 15 (≈$2.23) new-user voucher explicitly does not apply to K3. On the USD-priced side there is a promotional 'top up USD 5, get a USD 5 voucher' offer. Once credits are spent the platform is purely paid; no permanently free model is offered.

Official Moonshot AI developer platform for Kimi models. Offers pay-as-you-go access to Kimi K3, K2.7-Code, and K2.6 over an OpenAI-compatible API. One-off trial credits available upon verification. Details were checked against the provider's own page at platform.moonshot.cn.

After phone-number verification, new users automatically receive a RMB 15 (≈$2.23) trial voucher credited to their balance, usable directly to offset API call costs; the voucher typically has a 30-day validity and cannot be used for the latest models such as K3. Once used up or expired, it converts to pay-as-you-go.

Cohere provides enterprise-ready LLMs optimized for search, RAG, and agentic workflows. Its official API offers free Trial keys for development and evaluation, alongside high-performance production endpoints for models like Command R+ and the Aya multilingual series.

Trial key: 1,000 API calls/month (resets monthly, no card), Chat ~20 RPM, all models but non-production use only. ≈$2/mo assuming ~2K tokens/call at blended $1 per 1M tokens.

GitHub Models (models.github.ai) was an official model-inference platform by GitHub for prototyping. As of July 30, 2026, the service has been fully retired and the catalog/API are no longer available. Users are directed to Azure AI for production workloads.

Free with a GitHub account, no card; ~50 calls/day per higher-tier model (each model has own quota), 429s on heavy use; positioned for learning/prototyping. ~$1.5/mo equivalent at ~1K tokens/call.

Bytez is an AI model hosting platform that offers an OpenAI-compatible API for a wide range of models. It provides $1 in free credits that refresh every 4 weeks and uses a low-cost pay-as-you-go pricing model.

$1 free credits, refreshes every 4 weeks

Fireworks AI is a high-speed serverless inference platform for open-weights models (Llama, Qwen, DeepSeek). It provides an OpenAI-compatible API optimized for low latency and high throughput. Details were checked against the provider's own page at docs.fireworks.ai.

$1 free trial credit on sign-up.

Hyperbolic (hyperbolic.ai) is an open-access AI cloud providing high-performance serverless inference and GPU rentals. It offers an OpenAI-compatible API to 25+ open-source models including DeepSeek, Qwen, and Llama 3.1 405B. New users receive roughly $1 in trial credits upon phone verification.

~$1 trial credit on signup (phone verification required); occasional promotional credits ($15-$25 for hackathons/events); then PAYG from ~$0.0001/1K tokens. No long-term free flagship quota.

Hyperbolic (hyperbolic.xyz) is the API-specific endpoint for Hyperbolic's AI cloud. It provides serverless inference for over 25 open-source models through an OpenAI-compatible /v1 endpoint. The platform supports text, image, and audio generation with pay-as-you-go pricing.

~$1 trial credit for new users (DeepSeek V3, Llama 3.1 405B, Qwen); pay-as-you-go after it is used (Llama 405B ~$4/M, DeepSeek R1 ~$3/M tokens).

Nebius AI Studio is a European AI cloud offering an OpenAI-compatible API to 60+ open-source LLMs; it provides high-performance inference-as-a-service with European data residency and transparent pay-as-you-go pricing.

AI Studio: ~$1 trial credit for new accounts (early $100-era promos ended; startup program can grant far more); plus ongoing free API for select open models (Llama 3.3 70B, Qwen3-235B) with tier-based rate limits.

Nebius AI Studio is a European-based GPU cloud platform providing managed inference for top open-source LLMs like Llama 3.1 and DeepSeek R1. It leverages its own EU data centers to offer high-performance, privacy-compliant inference with transparent pay-as-you-go pricing tailored for both developers and enterprises.

AI Studio grants new accounts ~$1 one-time trial credit (startup/promo programs can grant far more); Nebius AI Cloud free-trial program suspended as of 2026-07-13; studio models otherwise pay-as-you-go.

OVHcloud AI Endpoints is an inference API from the established European cloud provider OVHcloud. It offers OpenAI-compatible access to open-weight models with anonymous and registered free tiers for developers.

AI Endpoints free tier: rate-limited calls with no/empty API key; registered users get higher limits. No token quota or rate numbers published - not quantifiable. Totals subject to policy change.

iFlytek Spark (Xunfei Xinghuo) is a large model platform providing access to the Spark LLM family. It offers a permanently free Spark Lite version and significant one-time token allowances for newer users. The API is OpenAI-compatible and supports various modalities including text, image, and voice.

Spark Lite: unlimited tokens, QPS 2, no expiry (permanently free - unquantifiable); each version ~2M free tokens for ~1 year; campaigns grant 100M Spark Max tokens. OpenAI-style /v1 endpoint, console API Password.

DeepSeek is a leading AI research lab providing frontier-class models (V4-Flash/Pro, R1) via an OpenAI-compatible API. It is known for industry-leading cost efficiency and specialized thinking/reasoning modes.

5M free tokens on sign-up; estimated value based on V4-Flash peak rates.

Novita AI is a developer-focused inference cloud offering an OpenAI-compatible API to models like DeepSeek, Qwen, and Llama. New users receive a $0.50 trial credit; service is otherwise pay-as-you-go with very low latency.

~$0.50 free credit for new users to test text/image APIs (~60 RPM); referral program gives $10 to each side up to $500 total (promotional); pay-as-you-go once credit is used.

Scaleway Generative APIs offer a serverless, OpenAI-compatible endpoint for deploying open LLMs. Hosted in Europe, it emphasizes data sovereignty and grants 1,000,000 free tokens to all customers to jumpstart development.

Value based on 1M input tokens at standard rate (approx €0.2).

Cloudflare Workers AI allows running AI models on Cloudflare's global network. It provides an OpenAI-compatible API for inference, embeddings, and image generation, billed in compute 'Neurons'. Details were checked against the provider's own page at developers.cloudflare.com.

10,000 Neurons per day free allowance.

Hugging Face is the leading hub for open-source AI, offering model hosting, datasets, and serverless inference. It provides a permanent free tier for basic CPU spaces and community GPU grants, with professional-grade Inference Endpoints available on a pay-as-you-go basis.

Official monthly inference credits: $0.10/mo for Free users, $2.00 for PRO, $2.00/seat Team/Enterprise; Serverless Inference free allowance for models under 10GB. Subject to change.

Nomic AI provides the Atlas Embedding API and open-weight models like nomic-embed. It offers a monthly free tier for its embedding API and is transitioning to focus on document AI for specialized industries.

Atlas embedding API free tier ~1M embedding tokens/month for evaluation and prototyping; ~$0.10/mo equivalent at Nomic embed pricing (~$0.10/1M tokens); paid beyond that.