Compare gateways
Compare AI gateways
Experiential beside OpenRouter, Kong, LiteLLM, Portkey, Cloudflare, Vercel and Helicone, feature by feature with sources.
At a glance
This page puts Experiential next to seven other AI gateways, one row per capability. We checked every competitor cell against that vendor's own docs or pricing page on 2026-11-19. "Not documented" means we could not find it there, not that it is missing.
| Experiential | OpenRouter | Kong AI Gateway | LiteLLM | Portkey | Cloudflare AI Gateway | Vercel AI Gateway | Helicone | |
|---|---|---|---|---|---|---|---|---|
| Hosting | Hosted | Hosted | Konnect hosted or self-hosted | Self-hosted, open source | Hosted, hybrid, or open-source gateway | Hosted | Hosted | Hosted or open source |
| OpenAI Chat and Responses | Both | Both | Both | Both | Both | Chat (compat), Responses via OpenAI pass-through | Both | Chat |
| Anthropic Messages | Native, every model | Yes | Native format, Anthropic only | Yes | Yes, any provider | Pass-through to Anthropic | Yes | Not documented |
Gemini generateContent | Native, every model | Not documented | Native format, Gemini and Vertex only | Yes | Not documented | Pass-through to Google | Not documented | Not documented |
| Embeddings | Yes | Yes | Yes | Yes | Yes | Pass-through | Yes | Not documented |
| Batches | Yes | Yes | Yes | Yes | Yes | Not documented | Batch jobs mentioned in pricing | Not documented |
| Price on tokens | List price, no markup | 5.5% fee on credit purchases (Standard) | No model resale; plan and request fees | Free (open source); Enterprise custom | Per request: Pro $49 a month for 100K, then $9 per 100K | 5% fee on Unified Billing credits | List price, no markup | 0% markup on credits |
| Bring your own key | No markup; first paid key needs Pro | No fee to $25K a month, then 5% | Your keys only | Your keys only | Your keys | Stored keys | No markup; $0.20 per 1K if fallback is off | Yes |
| Failover | Ordered waterfall; slow lanes demoted | Provider and model fallbacks | Load balancing with fallback (Enterprise plugin) | Fallbacks and retries | Fallbacks, retries, circuit breakers | Retries, model fallback | Provider and model fallbacks | Automatic failover |
| Zero data retention | Pro: org switch or per request; verdict on every response | Org setting and per request | Not documented | Not documented | Not documented | Unified Billing only, not BYOK | Per request free; team-wide $0.10 per 1K | Not documented |
| Spend limits | Per-key caps, budgets, tier rate limits | Per-key limits, guardrail budgets | Token budgets per consumer group | Per key, team and user | Per key and provider | Spend limit per gateway | Per team, project, key, member | Custom rate limits |
| Content guardrails | Standard pack or your own classifier, per key | Model, provider and moderation policies | Prompt guards, PII sanitizer, cloud safety | Content filters, PII masking | 50+ checks | Llama Guard, DLP | Not documented | Not documented |
| Response cache | None (provider prompt caching passed through) | Not documented | Semantic cache (Enterprise plugin) | Exact and semantic | Simple and semantic | Yes | None (prompt caching only) | Yes |
| Models | 980 listed, all callable | 600+ | 17 providers, your keys | 100+ providers | 3,000+ models, your keys | 14 providers on the compat endpoint | Not documented | 100+ |
| SSO | Not yet available | Enterprise | Enterprise | Enterprise | Enterprise | Not documented | Not documented | Enterprise (SAML) |
Where Experiential is better
- Four wire formats on one hosted key. OpenAI Chat, Responses, native Anthropic Messages and native Gemini
generateContentall reach every catalog model, plus embeddings and batches. Among the hosted gateways here, only Experiential documents all four. LiteLLM matches it, but you host it. See Models and the catalog. - Zero data retention you can verify per response. One org switch, or
provider: {"zdr": true}on a single request (OpenRouter's field shape). Unknown providers count as not-ZDR. It applies to every lane, your own keys included, on Pro with no per-request fee. Every completion carriesx-gateway-zdrandx-gateway-provider, and the list of zero-retention routes is public. Cloudflare's ZDR covers Unified Billing only, and Vercel charges for team-wide ZDR. See Data controls. - No-training routing on every plan.
require_no_trainingneeds no paid plan and applies to your own keys too. See Data controls. - Slow lanes get demoted, not just failed ones. A provider that answers slowly moves down the waterfall on its own, then recovers on its own. Failover never cuts you off to protect a cache hit. See Routing and failover.
- No markup on tokens, either lane. Credits pay list price. Your own key is billed by the provider, with no per-request fee and no volume threshold (a first paid-provider key needs Pro). OpenRouter charges 5% on BYOK above $25K a month, and Vercel charges $0.20 per 1K BYOK requests with fallback off. See Plans and pricing.
- A drop-in OpenRouter surface. It keeps the same
/api/v1paths,vendor/slugmodel ids,:freespellings, and/generation,/creditsand/keyreads. Moving off OpenRouter is a base URL and key change. See Compare: OpenRouter. - Cost in the response. Finished Chat, Responses and Messages replies carry
usage.costand theproviderthat served them, so billing a request needs no second lookup. - Guardrails with your own classifier. Per key, run the standard PII, secret and injection pack, or call your own HTTP endpoint. Prompts and completions are never stored by the guardrail. See Content guardrails.
curl -si https://api.experientiallabs.ai/v1/chat/completions \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "gpt-6-luna", "provider": {"zdr": true},
"messages": [{"role": "user", "content": "Hello"}]}' \
| grep -iE "^x-gateway-|^x-request-id|\"cost\""When another gateway may fit better
- You must self-host or run air-gapped. LiteLLM, Kong, Portkey's hybrid or open-source gateway, and Helicone's open-source build run in your network. Experiential is hosted.
- You already run Kong for API management. Keep it, and point its
ai-proxyupstream at Experiential (see Kong). - You need a response cache. Portkey, LiteLLM, Cloudflare and Helicone cache whole responses, and Kong offers a semantic cache. We pass provider prompt caching through but do not cache responses.
- You need SSO today. OpenRouter, Kong, LiteLLM, Portkey and Helicone offer SAML or OIDC on Enterprise plans. Ours is not yet available.
- Your app lives on Vercel or Cloudflare Workers. Their gateways plug into OIDC and Worker bindings with no key to manage.
Sources, checked 2026-11-19
- Experiential: /llms.txt ("Two lanes", "Zero data retention", "Models"), live model count from GET /api/models (
total: 980), public ZDR route list, AI data policy. - OpenRouter: pricing, FAQ, Messages, embeddings, batch API, guardrails, provider routing, ZDR.
- Kong: AI Gateway, ai-proxy plugin, pricing.
- LiteLLM: proxy, supported endpoints, caching, enterprise.
- Portkey: feature comparison, feature overview, AI gateway, universal API, batches. portkey.ai/pricing now redirects to a Palo Alto Networks contact form.
- Cloudflare: AI Gateway, pricing, Unified Billing, OpenAI-compatible endpoint, Anthropic.
- Vercel: AI Gateway, pricing, security and compliance.
- Helicone: AI Gateway, pricing. The pricing page states that Helicone is now part of Experiential Labs.