OpenRouter is genuinely zero-markup on inference tokens (what you pay matches each provider's published rate), but it charges a 5.5 percent fee on credit top-ups, and the routing layer adds roughly 25 to 40 ms of latency. The cost flips when a focused alternative covers the one feature you actually use: OSS self-hosters want LiteLLM, production teams want Portkey, edge-native teams want Cloudflare, and Vercel-hosted apps want the gateway already bundled with their plan.
Where alternatives win
Vercel AI Gateway is bundled with the Vercel plan you already pay for, ships hundreds of models with no token markup, and integrates tightly with the AI SDK that most Vercel-hosted Next.js apps already use.
Portkey open-sourced the full gateway under Apache 2.0 in March 2026 (1T+ tokens daily across 24,000+ orgs); the hosted Production tier at $49 monthly bundles prompts, observability, semantic caching, and guardrails that gateway-only OpenRouter leaves you to build.
LiteLLM is Apache 2 / MIT-licensed Python proxy with 100-plus provider integrations; self-hosted is the lowest absolute cost for engineering-heavy teams comfortable running Postgres plus Redis to back it.
Cloudflare AI Gateway is free up to 10,000 requests per day, runs on the same edge network as your Workers, and pays back the most for teams already inside the Cloudflare developer platform.
By Subrupt EditorialPublished Reviewed
LLM gateways became a category in 2023 once teams realized that hardcoding a single provider into apps creates two compounding fragilities: rate limits and outages on the inference side, and zero negotiation power on the cost side. The pattern that emerged is route every call through a gateway that handles caching, fallback, observability, and cost tracking, so the underlying provider mix can change without touching app code. OpenRouter took the marketplace approach with 300-plus models and pass-through pricing.
The OpenRouter wedge is breadth and zero inference markup. You hit one OpenAI-compatible endpoint, you can flip from Claude Sonnet to GPT-4o to Gemini in a config string, and the published token rates match what each provider charges directly. The friction sits two layers down: the 5.5 percent fee charged on credit top-ups, the routing latency that adds 25 to 40 ms over a direct call, and the gateway-only feature scope. Prompt management, evals, guardrails, and a real production dashboard are not in the box.
Four exit lanes arrive here. Vercel-hosted teams who can use a gateway already bundled with their existing Pro plan at no incremental token markup. Production teams who want bundled prompts, observability, semantic caching, and guardrails (Portkey now ships those Apache 2.0 self-hostable). Engineering-heavy teams with the DevOps capacity to run a self-hosted Python proxy at provider rates with no fee on top. Cloudflare-native teams whose stack already pays for Workers and gets edge gateway free up to 10K requests per day.
Quick map by stack. Already on Vercel: Vercel AI Gateway. Need bundled production features: Portkey. Have DevOps capacity and want zero managed cost: LiteLLM. Already on Cloudflare Workers: Cloudflare AI Gateway.
Affiliate disclosure: Subrupt earns a commission when you switch to a service through our recommendation links. This never changes the price you pay. We only recommend services where there's a real cost or feature advantage for you, and our picks are based on the data on this page, not on which programs pay the most.
Quick pick by use case
If you only have thirty seconds, find your situation below and skip to that pick.
Free up to 10K requests/day; Workers Paid at $5/mo; built into the same edge network as your app code.
Skip these picks if: If you actively rely on OpenRouter's 300-plus model marketplace breadth (especially the 25-plus free models for experimentation) or you have already invested in routing config across exotic providers, the 5.5 percent credit-purchase fee is doing real work and any pick below trades catalog reach for fixed pricing or a richer dashboard.
At a glance: OpenRouter alternatives
Quick comparison across pricing floor, best fit, and switching effort. Tap a row to jump to the full pick.
Best for Cloudflare Workers stacks at zero incremental cost
$5/mo + higher limits
Low
Feature comparison
Feature
Vercel AI Gateway
Portkey
LiteLLM
Cloudflare AI Gateway
Free tier
✓
✓
✓
✓
Self-hostable OSS
✗
✓
✓
✗
Provider failover routing
✓
✓
✓
✗
Prompt management
✗
✓
✗
✗
Guardrails
✗
✓
✓
✗
Semantic caching
✓
✓
✓
✓
Observability dashboard
✓
✓
~
✓
BYOK at list price
✓
✓
✓
✓
Provider catalog depth
hundreds
1,600+
100+
tens
Entry paid tier
$20/user
$49/mo
$50/user
$5/mo
Cost at your volume
Approximate cost per pick at typical requests/mo.
Pick
Hobby10,000 requests/mo
Production200,000 requests/mo
Scale1,000,000 requests/mo
Vercel AI Gateway
Free
$20/mo
$20/mo
Portkey
Free
$49/mo
$249/mo
LiteLLM
$50/mo
$50/mo
$50/mo
Cloudflare AI Gateway
Free
$5/mo
$200/mo
Modeled on top of $0 markup providers (token cost identical across all four). Numbers shown are gateway/platform fees only at each level; OpenRouter shown for reference assuming a $1,000 monthly inference spend (5.5 percent credit-purchase fee = $55/mo).
Vercel AI Gateway is what OpenRouter would look like if it were free for anyone already paying for Vercel. Hundreds of models behind one endpoint, zero markup on tokens, BYOK at provider list price, and tight integration with the AI SDK that Next.js teams already pull in.
The trade: Smaller curated catalog than OpenRouter's 300-plus marketplace, deeper friction outside the Vercel ecosystem, and the most powerful surfaces (custom reporting, provider analytics) are scoped to Pro and Enterprise plans.
The upside: No incremental gateway invoice on top of the Vercel plan you already pay for, no credit-purchase fee, and a real production case study: Zo Computer cut retry rate from 7.5 percent to 0.34 percent (a 20x reliability improvement) after switching.
“Moving to the gateway is just so ergonomic. We get references to model names, and then rely on you to do the correct implementations and handle the edge cases.”
“We didn't even know what we were missing until after we switched to Vercel's AI Gateway. The revelation came through the numbers.”
Strengths
+Bundled with Vercel Pro at $20/user
+Zero markup on tokens; BYOK at list price
+Hundreds of models, OpenAI-compatible endpoint
+Tight AI SDK + Next.js integration
Trade-offs
−Smaller catalog than OpenRouter's marketplace
−Best fit only for Vercel-hosted stacks
−Custom reporting API gated behind Pro/Enterprise
Hobby
$5 free credit/mo
Pro
$20/user/mo (bundled with Vercel Pro)
Enterprise
Custom + SAML SSO
Pricing verified
2026-05-10
Migration steps
Confirm your Vercel plan tier (Hobby covers free credit; Pro at $20 per user adds load balancing and budgets).
Enable AI Gateway in the Vercel dashboard and pick which models to route.
Swap OpenAI/Anthropic base URLs for the Vercel AI Gateway endpoint (or upgrade your AI SDK to v5/v6 if you have not).
Add provider keys via BYOK to keep direct contracts active and avoid provider-side billing surprises.
Cancel the OpenRouter Pro credit minimum once Vercel covers your routing surface.
Not for: Vercel AI Gateway is the wrong fit for teams not hosted on Vercel or for any team that needs OpenRouter's full 300-plus marketplace catalog; LiteLLM, Portkey, or staying on OpenRouter cover those better.
Portkey is the production gateway that ships everything OpenRouter intentionally leaves out. The full gateway is now Apache 2.0 self-hostable (open-sourced in March 2026, handles 1T-plus tokens daily across 24,000-plus orgs), and the hosted Production tier at $49 monthly covers 100,000 logs with prompts, semantic caching, retries, fallbacks, RBAC, alerts, and LLM guardrails.
The trade: Slightly smaller model catalog than OpenRouter's marketplace breadth, the Production tier 100K log limit can compound for high-volume teams pushing to the $249 Pro tier, and the AI-native features add roughly 20 to 40 ms latency on top of the gateway hop.
The upside: One platform replaces gateway-plus-Langfuse-plus-prompt-tool stacks, and the Apache 2.0 self-host route gives a credible exit if hosted pricing ever bites.
Strengths
+Apache 2.0 self-hostable since March 2026
+Bundled gateway + prompts + guardrails + caching
+Free Developer tier at 10K logs/mo
+SOC 2 Type II + HIPAA on Enterprise
Trade-offs
−Catalog smaller than OpenRouter's marketplace
−Production tier 100K log limit pushes scaling teams to Pro
−AI-native features add latency on top of routing
Open Source
$0 self-host (Apache 2.0)
Developer Free
$0/mo, 10K logs
Production
$49/mo, 100K logs + guardrails
Pro
$249/mo, 1M logs + RBAC
Pricing verified
2026-05-10
Migration steps
Decide self-host (Apache 2.0 OSS gateway) or hosted (Developer Free, then Production at $49).
Configure Virtual Keys for each provider so Portkey owns rotation and rate limits.
Swap your app's API base URL to the Portkey gateway and add the project ID.
Wire the prompt registry and observability dashboard for at least one production route before broad rollout.
Cancel OpenRouter once Portkey's catalog covers the providers you actually use.
Not for: Portkey is the wrong fit for teams who need OpenRouter's 300-plus model marketplace breadth or for solo devs who want the lightest possible gateway with no platform features attached; OpenRouter or LiteLLM fit those better.
LiteLLM is the Apache 2 / MIT-licensed Python proxy that translates 100-plus provider APIs into one OpenAI Chat Completions schema. Self-hosted runs at provider rates with no markup; Cloud Pro at $50 per user monthly adds cost tracking, budgets, virtual keys, and team management.
The trade: You operate the proxy. Postgres for spend logs and API keys, Redis for cache and rate-limit counters, your own monitoring; honest community signal puts the all-in operational cost between $500 and $3,000 monthly once DevOps time is counted, and recent Hacker News threads flag bug-prone enterprise feature surface.
The upside: Zero managed-vendor markup, full control over routing logic and data residency, and a credible escape hatch when any hosted gateway repricing happens upstream.
Strengths
+Apache 2 / MIT licensed, self-hostable
+100-plus provider integrations
+Cost tracking and budgets on Cloud Pro
+Virtual API keys for team usage
Trade-offs
−Operational overhead from Postgres + Redis + monitoring
−Enterprise feature surface has been bug-prone in 2025-2026
−Routing intelligence shallower than OpenRouter's auto-fallback
Open Source
$0 self-host (Apache 2 / MIT)
Cloud Free
Limited request volume
Cloud Pro
$50/user/mo
Enterprise
Custom + SOC 2 + HIPAA
Pricing verified
2026-05-10
Migration steps
Decide self-host (Docker plus Postgres plus Redis) or LiteLLM Cloud (Free tier first).
Configure provider API keys and routing rules in config.yaml or Cloud UI.
Switch your app's OpenAI base URL to the LiteLLM proxy URL.
Validate cost tracking, virtual keys per team, and at least one fallback route under production load.
Cancel OpenRouter Pro once LiteLLM covers your routing surface and the operational cost is below the markup it replaces.
Not for: LiteLLM is the wrong fit for teams without DevOps capacity to operate Postgres plus Redis, or those who need OpenRouter's full marketplace breadth; OpenRouter, Portkey hosted, or Vercel AI Gateway cover those better.
Cloudflare AI Gateway is free up to 10,000 requests per day with caching, analytics, rate limits, and analytics across OpenAI, Anthropic, Google, and Workers AI. Workers Paid at $5 monthly raises request limits; Workers Standard at $200 monthly bundles 50M Workers requests with unlimited AI Gateway calls.
The trade: Smaller model catalog (focused on the headline providers plus Workers AI), thinner prompt management UI than Portkey, no provider failover routing logic in the gateway itself, and TrueFoundry has documented one scaling startup hitting unpredictable logging costs at very high request volume.
The upside: For Cloudflare-native teams the gateway is essentially free incremental capability on the same edge network, and Cloudflare's own internal AI engineering team routes 20M-plus requests per month through it, a credible self-dogfood signal.
Strengths
+Free up to 10K requests/day
+Caching, analytics, rate limits built in
+Same edge network as your Workers
+Bundled into Workers Paid at $5/mo
Trade-offs
−Smaller catalog than OpenRouter's marketplace
−No provider failover routing logic in-gateway
−Logging costs can compound at very high volume
−Best fit only for Cloudflare-friendly stacks
Free
10K requests/day
Workers Paid
$5/mo + higher limits
Workers Standard
$200/mo + unlimited AI Gateway
Enterprise
Custom + dedicated regions
Pricing verified
2026-05-10
Migration steps
Sign up for Cloudflare (Free or Workers Paid at $5 if your app already runs on Workers).
Create the AI Gateway project in the Cloudflare dashboard.
Swap your OpenAI/Anthropic/Google base URL for the AI Gateway endpoint.
Configure caching, rate limits, and budgets to keep logging costs predictable at volume.
Cancel OpenRouter once Cloudflare covers your routing surface.
Not for: Cloudflare AI Gateway is the wrong fit for teams who need OpenRouter's full 300-plus marketplace catalog, want richer prompt management out of the box, or run on a non-Cloudflare stack; OpenRouter, Portkey, or Vercel AI Gateway fit those better.
Paid plans from $5.00/mo
When to stay with OpenRouter
Stay with OpenRouter if your app already routes across the 300+ model marketplace with auto-fallback wired up, the 25-plus free models cover an experimentation surface you genuinely use, or your monthly inference spend is small enough that the 5.5 percent credit-purchase fee rounds to noise. The picks below address self-hosted OSS proxy, production-grade observability bundle, edge-native zero-cost routing, and Vercel-stack-native app delivery.
Picks split across four dimensions readers actually decide on: deployment model (managed-cloud vs OSS-self-hosted vs frontend-stack-native vs edge-native), feature scope (gateway-only vs gateway-plus-observability), pricing model (markup-on-tokens vs flat-monthly vs request-tier vs bundled-with-platform), and provider catalog depth.
Pricing pulled from each vendor's site on the review date and cross-checked against catalog pricing-history annotations. We score on cost-at-volume for representative LLM apps (10M tokens monthly across mixed providers; 200K and 1M monthly request shapes), feature breadth, catalog depth, and operational lift to migrate. We weight against tools whose advertised free tier excludes essential features that quickly push teams to paid, and we drop honorable-mention picks (TrueFoundry, Vellum) when their entry tier ($300 to $500 monthly) sits above the realistic switcher's cost ceiling.
Update history2 updates
Initial published version with 5 picks.
Backfilled to Stage 2 schema with structured verdict, 4-paragraph intro, Quick Verdict, Feature Matrix, Usage Cost Table, sourced testimonial, and per-pick author ratings. Corrected stale 5 percent inference-markup framing: OpenRouter actually passes through provider rates at zero markup and charges a 5.5 percent fee on credit top-ups instead. Replaced TrueFoundry and Vellum picks (MLOps-platform tier, audience-misfit for gateway-only switchers) with Vercel AI Gateway. Portkey now Apache 2.0 self-hostable as of March 2026.
Frequently asked questions about OpenRouter alternatives
Does OpenRouter actually charge a 5 percent markup on tokens?
No, this is a stale framing. OpenRouter passes through provider rates at zero markup on inference tokens; Claude Sonnet 4.5 costs the same on OpenRouter as it does direct from Anthropic. The fee that exists is on credit purchases: a $100 top-up yields roughly 94.50 of inference power (5.5 percent fee, with a small minimum). On the routing path itself, OpenRouter adds roughly 25 to 40 ms of latency over a direct provider call.
Why use a gateway at all instead of calling provider APIs directly?
Three reasons. Provider failover when an OpenAI or Anthropic outage hits, automatic retry on a backup. Cost optimization through routing to the cheaper provider when the quality difference is acceptable. Observability through centralized logging, cost tracking, and prompt versioning across providers. The cost (markup, subscription, or DevOps time) typically pays back when multi-provider usage runs above $1,000 monthly. Below that, direct API calls are usually cheaper end-to-end.
How does the OpenRouter credit-purchase fee compare to flat-fee gateways?
Math at $1,000 monthly inference spend on OpenRouter: roughly $55 in credit-purchase fees per month, with no platform tooling included. Portkey Production at $49 monthly covers 100K logs with bundled prompts, observability, semantic caching, and guardrails. Vercel AI Gateway is $20 per user on the Vercel Pro plan you may already pay for. Cloudflare AI Gateway is free up to 10K requests per day. The crossover where flat-fee beats credit-purchase fee is around $400 to $1,000 monthly inference spend, earlier if your app already runs on Vercel or Cloudflare.
Should I self-host LiteLLM or use a managed gateway?
Self-hosting works for engineering-heavy teams with DevOps capacity and inference spend high enough that the markup or subscription savings clear the operational overhead. Honest community signal puts the all-in cost of running LiteLLM at $500 to $3,000 monthly once Postgres, Redis, monitoring, and on-call time are counted. Managed gateway works for solo devs and teams under 50 engineers where DevOps time is more expensive than the markup it replaces. The biggest hidden cost of self-hosting is observability infrastructure that managed gateways include.
Can I use multiple gateways together?
Yes, sometimes. Pattern 1: OpenRouter for marketplace breadth as the underlying provider plus Portkey or LiteLLM as the observability layer in front. Pattern 2: Cloudflare AI Gateway for caching plus a richer gateway for routing logic. Pattern 3: direct provider APIs for high-volume routes plus a gateway for low-volume routes. Most teams pick one primary gateway and use direct provider calls as the escape hatch for the few routes that need it.
Ready to switch?
Our top OpenRouter alternative: Vercel AI Gateway
Vercel AI Gateway is bundled with the Vercel plan you already pay for, ships hundreds of models with no token markup, and integrates tightly with the AI SDK that most Vercel-hosted Next.js apps already use.
The team behind subrupt.com. We track subscriptions, surface cheaper alternatives, and publish comparisons where the score formula is on the page so you can recompute it yourself. We do not claim 30,000 hours of testing. What we claim is live pricing from our database, a transparent composite score, and honest savings math against a category baseline.
Get notified of price drops for OpenRouter
We'll email you when OpenRouter or its alternatives lower their prices.
Track OpenRouter and find more savings
Add OpenRouter to your dashboard to monitor spending and discover even more alternatives.