Directory model IDQwen/Qwen3.8-27B
| Listed field | Novita | Cerebras |
|---|---|---|
| Input / 1M tokens | $0.42 | $0.99 |
| Output / 1M tokens | $3.00 | $1.49 |
| Context tokens | 1,000,000 | 65,536 |
| Tools | Yes | Yes |
| Structured | No | Yes |
Free API. Trial credits. Know the difference.
Compare access conditions before you create an account. Shortlist up to three routes to see their limits and setup side by side, then bookmark or share the comparison.
Start with the connection your app needs.
“No signup”, “no credit card” and “no API key” describe different conditions. Check the interface, account and billing before choosing a route.
I have an app asking for an API URL and keyHosted OpenAI-compatible API
Use a provider’s API base URL, its key and a supported model ID. The ten hosted routes below compare free allowances, trials and paid access. Check each route’s account setup and expiry; an offer without a credit card can still require signup.
For ChinaGenAI’s automatic route, check key access, then copy the connection settings. Its short-text limit is 256 output tokens. Local response inspection works without signup, a key or a model request.
Already have a response? Inspect it locallyI’m building a web app with JavaScriptBrowser SDK, such as Puter.js
Puter.js uses calls such as puter.ai.chat(). Under its User-Pays model, users sign in with a Puter account and cover their own usage. A monthly free allowance can cover some requests; running out can prompt an upgrade.
A no-key SDK tutorial does not establish free access to a separate endpoint. Puter’s current quota documentation requires a paid plan for /puterai/openai/v1/*; a free account receives 402 subscription_required.
I want to run a model on my own computerLocal server, such as Ollama
Ollama’s local API does not require authentication. Its OpenAI-compatible examples use http://localhost:11434/v1/; a client may require a placeholder key that the local server ignores. Install Ollama and download a suitable local model first.
Your device supplies the compute. localhost means the machine running the client; software hosted elsewhere needs a reachable server address. Ollama cloud models have separate sign-in and API-key requirements.
Sources checked . These are documented access conditions; no provider call or local installation was tested.
Compare a model across hosts
Qwen and DeepSeek: listed prices, context and request features
Context limits, prices and exposed features can differ between hosts. These are Hugging Face directory model IDs, not copy-ready IDs for direct host requests. For a direct connection, get the request model ID from that host’s documentation or your account’s model list.
Directory model ID
Directory model IDdeepseek-ai/DeepSeek-V4.1-Flash
| Listed field | Novita | Baseten |
|---|---|---|
| Input / 1M tokens | $0.30 | $0.30 |
| Output / 1M tokens | $1.20 | $1.20 |
| Context tokens | 1,048,576 | 1,048,576 |
| Tools | Yes | Yes |
| Structured | No | Yes |
Source: Hugging Face’s model/provider directory, checked . “Tools” and “Structured” reproduce that directory’s columns. A “No” does not mean the model cannot write JSON; check the host’s documented request options.
Context is a listed window, not your remaining input budget after output reservation. Check the host’s current terms for cache pricing, account credits and other charges.
These are dated directory listings, not live tests, free-access offers or models you can select through ChinaGenAI. No speed or quality ranking is inferred. Download the source-backed snapshot.
Preparing comparison controls… · 10 of 10 routes · Checked 2026-10-21
Your shortlist 0/3
Select up to three routes below. Filtering keeps your shortlist.
ChinaGenAI
Automatic route
Free allowance / variants
Limited free access
20 attempts per network or verified account per UTC day; 48-hour keys. Availability must be checked.
Account setup and use case
Site key; free model label. One active request per key, two across the site; 256 output tokens.
Short text requests and learning the Chat Completions format.
Source review: · Access notes only; no live call
OpenRouter
Multi-provider router
Free allowance / variants
Specific free model variants
Only actual free catalog entries qualify. Rate limits and availability apply; paid variants have separate pricing.
Account setup and use case
OpenRouter account and key; exact model ID from its current catalog.
Comparing model choices through one provider connection.
Source review: · Access notes only; no live call
Groq
Direct inference
Free allowance / variants
Model-specific free-plan limits
Free-plan request and token limits vary by model. Check the limits for your account before a larger workload.
Account setup and use case
Groq account and API key; OpenAI-compatible API with documented differences.
A small text prototype with an explicit request/token allowance.
Source review: · Access notes only; no live call
Google Gemini API
Direct inference
Free allowance / variants
Free tier for selected models
Eligibility depends on model, region and account. Free-tier content can be used to improve Google products; paid-tier terms differ.
Account setup and use case
Google project and Gemini API key; use the documented OpenAI compatibility endpoint.
Existing OpenAI SDK code that needs a Gemini connection.
Source review: · Access notes only; no live call
Cloudflare Workers AI
Cloud platform
Free allowance / variants
Daily allocation for eligible models
10,000 Neurons per day. Some models require paid billing even with an allocation. Workers Paid charges for usage beyond the allocation.
Account setup and use case
Cloudflare account, account ID and API token; model-specific compatibility.
A Worker or REST integration with a documented daily compute allocation.
Source review: · Access notes only; no live call
Vercel AI Gateway
Multi-provider router
Free allowance / variants
$5 monthly free-tier credits
Free credits start with the first AI Gateway request and apply only to eligible models with lower limits. Buying credits moves the team to the paid tier and ends the monthly free credit.
Account setup and use case
Vercel team and AI Gateway key; check the eligible-model catalog and current account requirements.
An existing AI SDK or OpenAI-compatible application.
Source review: · Access notes only; no live call
SiliconFlow (international)
Multi-model inference
Limited trial
Introductory credits; metered API
The international pricing page advertises $1 introductory credits. Treat this as an account offer; do not assume the .cn service has identical terms.
Account setup and use case
SiliconFlow account and key. Check the international endpoint, model ID and billing in your account.
Exploring Chinese open models with published per-model pricing.
Source review: · Access notes only; no live call
Cerebras
Direct inference
Limited trial
$5 trial; expires after 30 days
A verified payment method is required to activate the trial. There is no permanently renewing free tier in the current documentation.
Account setup and use case
Cerebras account, verified payment method and API key. Access pauses when trial credits expire or run out.
A bounded evaluation before choosing a paid inference plan.
Source review: · Access notes only; no live call
Hugging Face Inference Providers
Multi-provider router
Paid / subscription
Credits purchase or subscription
Free accounts currently receive no included monthly inference credits. PRO and organization credits are subscription benefits, not a free API offer.
Account setup and use case
Hugging Face account and token; routed billing or your own provider credentials have different billing paths.
Finding which provider serves a model and comparing documented capabilities.
Source review: · Access notes only; no live call
DeepSeek
Direct inference
Paid / subscription
Metered API
Free consumer chat and a paid API account are separate. Check current input/output prices and model aliases.
Account setup and use case
DeepSeek account and provider key; current model ID and thinking-mode rules matter.
Using a named DeepSeek model with its provider-specific features.
Source review: · Access notes only; no live call
What this comparison checks
These are access routes, not ten models available through ChinaGenAI. A free allowance, one-time trial, subscription benefit and free consumer chat are different offers. Sources and review dates are attached to each route. No latency benchmark or quality ranking is implied.
Check model eligibility, country availability, data-use terms, account verification, expiry and paid overage on the provider’s current page. We do not request provider credentials here. Links are ordinary editorial links with no referral identifiers.
Need Qwen, Kimi or Doubao? Browse the model-specific connection guides. Have a response already? Inspect it locally without another model call.