OpenRouter lists 311 models as of May 2026; the standout additions from the current lineup are GPT-5.5 and Claude Opus 4.7.
Model Coverage & Stability
OpenRouter is not a curated list — it’s a firehose. Every major provider’s latest release lands here within hours. The data shows a 99.9% uptime for the gateway itself, but that’s the aggregate. Per-model availability is the real story. During the May 2026 Claude Opus 4.7 launch, OpenRouter hit rate limits on Anthropic’s side within 90 minutes. The gateway stayed up; the model showed 503s for 12 hours. Bottom line: The gateway is reliable, but individual models inherit their upstream provider’s stability. Key takeaways:
- 311 models total; the newest (GPT-5.5, Opus 4.7, Gemini 3.1 Pro) get added same-day
- During provider incidents, OpenRouter shows degraded status per-model, not globally
- No model-specific SLA — you get the upstream provider’s uptime, not OpenRouter’s 99.9%
Context Window & Speed Benchmarks
Max tokens across all models: 200,000. That’s the ceiling — DeepSeek V4 and Claude Sonnet 4.6 hit it; GPT-5.5 caps at 128K. Latency is 293ms average with a 5.0/5 speed rating. That’s fast, but misleading. The 293ms is median across all models. Actual performance varies by model weight and provider peering:
| Model | Context Window | Typical Latency (p50) | Rate Limited? |
|---|---|---|---|
| GPT-5.5 | 128K | 180ms | Yes — 500 RPM |
| Claude Opus 4.7 | 200K | 420ms | Yes — 100 RPM |
| Gemini 3.1 Pro | 200K | 250ms | No (pay-per-token) |
| DeepSeek V4 | 200K | 210ms | No |
| Llama 4 | 128K | 150ms | No |
| Mistral Large | 128K | 200ms | No |
| OpenRouter’s speed advantage comes from routing — it picks the fastest available provider for each model. During congestion, it falls back to a secondary provider automatically. | |||
| Key takeaways: |
- 200K context is available on Claude Opus 4.7, Gemini 3.1 Pro, and DeepSeek V4
- GPT-5.5 and Llama 4 are capped at 128K
- Latency varies 2x between models; Gemini 3.1 Pro is the best balance of context and speed
Pricing — The Catch
OpenRouter charges official price + markup fee. The markup is displayed per-model in the dashboard, typically 10-30% above base cost. There are no bulk discounts, no prepaid plans, no pay-as-you-go tiers.
| Pricing Dimension | OpenRouter |
|---|---|
| Base model cost | Official provider price + 10-30% markup |
| Free trial | Yes — $1 credit on signup |
| Monthly subscription | $0 (no plans) |
| Payment methods | Credit card (Stripe), USDT |
| Bulk discount | None |
| Minimum recharge | None specified |
| For a Chinese developer running 1M tokens/day on GPT-5.5, that markup adds roughly $30-50/month over direct API access. The trade-off is zero VPN overhead and access to models that would otherwise require separate accounts. | |
| Key takeaways: |
- Markup is transparent but unavoidable; no volume pricing exists
- Free $1 credit is enough for ~50,000 GPT-5.5 input tokens
- USDT payments work; no CNY option
Pros & Cons
Pros:
- 311 models — most full selection of any gateway
- Transparent per-model markup shown before you send a request
- 99.9% gateway uptime with automatic provider failover
- Supports credit card and USDT payments Cons:
- Most expensive option — official price + 10-30% markup, no discounts
- TLS handshake issues on China Telecom routes (per FAQ)
- No CNY billing, no Alipay/WeChat Pay
- Model-specific availability can drop during upstream provider incidents
Verdict
OpenRouter is the best choice if you need access to every major model without managing multiple API keys. The 293ms latency and 99.9% uptime are real. But you pay for that convenience — the markup adds up, and there’s no path to lower rates at scale. For Chinese developers: the TLS flakiness on China Telecom is a real issue. Test with a small workload first. If the handshake fails consistently, OpenRouter is not usable without a proxy — making it redundant with direct API access. If you need the newest models immediately (GPT-5.5, Claude Opus 4.7) and the markup is acceptable, OpenRouter is the best gateway. If you’re optimizing for cost or need CNY payments, look at API2D or Aihubmix.
FAQ
Is OpenRouter accessible from mainland China without a VPN?
The API endpoint is reachable from most Chinese ISPs, but TLS handshakes can be flaky on China Telecom routes. Test with a single request before committing.
Can I pay for OpenRouter in CNY?
No. Only USD via credit card or USDT. For CNY invoicing, use API2D or Aihubmix.
What models does OpenRouter support?
311 models as of May 2026, including GPT-5.5, GPT-4o, Claude Opus 4.7, Claude Sonnet 4.6, Gemini 3.1 Pro, Llama 4, DeepSeek V4, and Mistral Large.
Is my API key safe on OpenRouter?
Keys are hashed at rest (bcrypt), with support for per-project scoping, expiration dates, and IP allow-lists.
Which model has the best speed-to-context ratio?
Gemini 3.1 Pro — 200K context at 250ms latency with no rate limits. DeepSeek V4 is close at 210ms but has slightly higher per-token cost.
Data provenance: any figures from hands-on checks are author-reported and tested on the recorded date. Independent verification is unavailable unless a source is linked.
