🇺🇸 USA · OpenRouter
Status: 🟩 COMPLETE 🟦 LIVING Last updated: 2026-06-26 Plain-English tagline: One API key, hundreds of AI models. OpenRouter aggregates models from OpenAI, Anthropic, Google, Mistral, Meta, Cohere, xAI, plus open-weight and free models — single unified API, single bill. The “Stripe of AI APIs.”
Front-matter facts
| Field | Value |
|---|---|
| Vendor | OpenRouter Inc (USA) |
| Country / origin | 🇺🇸 USA |
| Recommended for Australian users? | ✅ Yes — fully accessible; AUS card accepted |
| Privacy summary | Standard developer terms; per-model privacy posture varies (OpenRouter forwards to underlying provider) |
| Free tier | Yes — some free models always available; pre-paid credits for paid models |
| Paid tiers | Pay-per-token; OpenRouter takes a small markup over direct provider pricing |
| First released | 2023 |
| Last reviewed | 2026-06-26 |
| Official site | https://openrouter.ai |
What it is
OpenRouter is an AI API aggregator — instead of getting separate API keys from OpenAI + Anthropic + Google + Mistral + 20 others, you get one OpenRouter key that works with hundreds of models from many providers via a unified API.
What’s accessible via OpenRouter:
- OpenAI (GPT family)
- Anthropic Claude (full lineup)
- Google Gemini (full lineup)
- xAI Grok
- Meta Llama (open weights)
- Mistral (closed + open)
- Cohere (Command R / R+)
- Mistral, Phi, Qwen, DeepSeek, Yi (open weights hosted via Together/Fireworks etc.)
- Free models — some providers offer free tier through OpenRouter (Llama variants, Mistral Nemo, etc.)
- Plus 100+ others
Why developers love it:
- One key, one bill — vs juggling 5+ provider relationships
- Automatic fallback — configure preferred model order; OpenRouter routes
- Cost transparency — see per-call pricing
- Quick model experimentation — try a new model without separate sign-up
- Free models for development — develop without paying
Why not OpenRouter:
- Slight per-token markup vs direct provider APIs (~5-10%)
- One more layer of trust (your data flows through OpenRouter to underlying provider)
- For maximum reliability at scale, direct provider relationships preferred
What you’d use it for
- Prototyping across many models — try GPT vs Claude vs Gemini vs Llama with one key
- Building multi-model apps — let users pick model, you bill OpenRouter
- Cost optimisation — pick cheapest model that meets quality bar; switch easily
- Free-tier development — use OpenRouter’s free model tier for dev work
- Simplifying provider management — one API key in your env vars
- Personal projects — easier than multiple provider accounts
How to use from Australia
- Sign up at
openrouter.ai. Free credit on sign-up. - Get API key from dashboard
- Use OpenAI-compatible API:
from openai import OpenAI client = OpenAI( api_key="...", base_url="https://openrouter.ai/api/v1" ) response = client.chat.completions.create( model="anthropic/claude-sonnet-4.6", # or openai/gpt-5, google/gemini-3-pro, etc. messages=[{"role": "user", "content": "Hello"}] ) - Browse openrouter.ai/models for the full catalog + pricing
- AUS card accepted
What it costs
Pay-per-token (per model)
- OpenRouter charges underlying provider rate + ~5-10% markup
- Some models have FREE tier via OpenRouter (Llama 4 70B free, Mistral free variants, others)
- Per-model pricing transparent at openrouter.ai/models
Pre-paid credits
- Load credits, OpenRouter draws from them
- AUS card accepted
- No subscription required
Free models
- Some Llama / Mistral / Phi variants have free tier (rate-limited)
- Good enough for dev / small projects
How it compares to alternatives
| Aspect | OpenRouter | Direct provider APIs | LiteLLM | AWS Bedrock |
|---|---|---|---|---|
| Multi-model with one key | Yes | No (per provider) | Self-hosted router | Yes (within AWS) |
| Model catalog | Vast (hundreds) | One per | Self-configured | AWS curated |
| Markup vs direct | ~5-10% | None (cheapest) | None (self-hosted) | Some |
| Free models for development | Yes | Vendor-specific free tiers | N/A | Limited |
| Enterprise contracts | Limited | Direct provider best | Self-hosted | Best for AWS shops |
| AUS data residency | Inherited from underlying provider | Provider-dependent | Self-hosted controlled | Yes (Sydney) |
| Best for | Prototyping + multi-model apps | Production at scale | Self-hosted multi-model | AWS shops |
For prototyping / multi-model apps, OpenRouter is hard to beat. For production at scale with enterprise SLAs, direct provider or cloud-vendor (Bedrock / Vertex / Azure OpenAI) often better.
Privacy / data handling
- OpenRouter forwards your request to the underlying provider
- Privacy posture is layered: OpenRouter’s terms + the underlying provider’s terms both apply
- Most underlying providers don’t train on inputs via OpenRouter (same as direct API access)
- For sensitive workloads, verify both layers and consider direct enterprise contracts
Recent changes
- 2026: Catalog expanded to ~300+ models
- 2025: Free model tier expanded
- 2024: Major growth phase
- 2023: Initial launch
Gotchas
- Markup is small but real — for high-volume production, direct provider access can save a few percent
- Free models are rate-limited — fine for dev; production needs paid tier
- Latency adds a hop — OpenRouter → provider → back; usually negligible (~50-100ms extra) but matters for real-time work
- For enterprise / compliance, direct provider or cloud-vendor relationships often required
- Reliability — if OpenRouter has an outage, your app is affected; consider failover to direct providers for critical apps
- Model availability changes — OpenRouter adds / removes models periodically; subscribe to their changelog
See also
- LiteLLM 🟥 — self-hosted multi-provider router
- Portkey 🟥
- Helicone 🟥
- Claude API overview 🟩 🟦
- OpenAI API 🟩 🟦
- Google AI Studio 🟩 🟦
- AWS Bedrock 🟩 🟦
- Vertex AI 🟩 🟦
- Together AI 🟩 🟦
- Hugging Face 🟩 🟦
- which-ai-for-which-job.md 🟩 🟦