LiteLLM alternatives for LLM gateways and model access
Portkey is the closest LiteLLM alternative: an MIT-licensed gateway you can self-host, now sold by Palo Alto Networks. For a hosted router with no servers, pick OpenRouter at a 5.5% fee on credits. Cloudflare AI Gateway is free at its core, Kong suits API teams, and Novita replaces providers, not the gateway.
LiteLLM alternatives at a glance
Five tools, each matched to one reason teams move off LiteLLM. LiteLLM sits on the first row for reference. Every price cell carries the month it was checked on the vendor's own pricing page.
| Alternative | Best for | Cheapest paid plan | Free plan or trial |
|---|---|---|---|
| LiteLLM (for reference) | A free, self-hosted gateway with keys and budgets per team | Enterprise, price on request (Nov 2026) | Open Source: $0, self-hosted; 30-day Enterprise trial license (Nov 2026) |
| Portkey | A self-hostable gateway with a vendor-run control plane | Price on request through Palo Alto Networks (Nov 2026) | MIT-licensed open-source gateway, $0 to self-host (Nov 2026) |
| OpenRouter | One hosted API for many models, no servers to run | Standard, pay as you go: 5.5% fee on credit purchases (Nov 2026) | Free: 50 requests a day on free models (Nov 2026) |
| Cloudflare AI Gateway | Caching, rate limits and logs at the edge for almost nothing | Workers Paid, $5 a month minimum (Nov 2026) | Core gateway features free on Workers Free (Nov 2026) |
| Kong AI Gateway | API platform teams that already run Kong | Konnect Plus, from $25 a month plus usage (Nov 2026) | 30-day trial, no credit card (Nov 2026) |
| Novita | Cheap open-model tokens from one provider | Pay per token, e.g. Llama 3.3 70B at $0.135 in, $0.40 out per million (Nov 2026) | Free to start, no free credits stated (Nov 2026) |
Portkey and Kong are gateways in the same sense LiteLLM is. OpenRouter and Cloudflare move the gateway onto someone else's servers. Novita is a model host, so it removes the need for a gateway only if one provider covers every model you call.
Why teams leave LiteLLM
The first reason is that you run it. LiteLLM is free to self-host, but its production guide calls Postgres the primary supported database and asks for Redis 7.0 or newer once you run more than one proxy instance. It suggests starting load tests at 1 vCPU and 4Gi of memory per worker, so an 8-worker setup begins around 8 vCPU and 32Gi. That bill and the on-call rota land on your platform team, not on BerriAI.
The second is the enterprise line. SSO, SCIM, JWT auth, audit logs and tag-based spend tracking sit in the Enterprise license, and litellm.ai lists no price for it (Nov 2026). A regulated team that needs those features on day one can't budget without a sales call, though the 30-day trial license arrives without one.
The third is supply-chain exposure. LiteLLM's March 2026 security update covered a compromised PyPI release; the Docker image was unaffected, but any team installing from PyPI now has to pin versions and use signed images. Some teams would rather a vendor carry that risk.
Last, the defaults need work. The request timeout defaults to 6,000 seconds, verbose logging is on unless you turn it off, and the production guide lists more than a dozen settings to change before live traffic. A hosted gateway ships those choices made.
Which LiteLLM alternative fits your work
LiteLLM does three jobs at once: it routes calls to 140+ providers in one format, it governs who can spend what, and it logs everything. Most teams that leave care most about one of those. Find yours below.
| Job | Best pick | Runner-up | Deciding fact |
|---|---|---|---|
| Self-hosted gateway with a managed dashboard | Portkey | Kong AI Gateway | Portkey's gateway is MIT-licensed; its hosted control plane is now quoted by Palo Alto Networks |
| One API for many models with no infrastructure | OpenRouter | Novita | OpenRouter charges token list price plus 5.5% on credit purchases |
| Caching, rate limiting and logs for the lowest cost | Cloudflare AI Gateway | Portkey | Cloudflare's analytics, caching and rate limiting are free |
| LLM traffic governed with your other APIs | Kong AI Gateway | Portkey | Kong Plus includes 1 million AI gateway requests a month, then $200 per extra million |
| Cheaper open-model inference | Novita | OpenRouter | Novita lists DeepSeek R1 0528 at $0.70 in and $2.50 out per million tokens |
Portkey, the LiteLLM alternative for a managed control plane
Portkey is the closest match to LiteLLM on this page. Its open-source gateway on GitHub is MIT-licensed and covers 45+ providers behind one API, with fallbacks on chosen error codes, retries up to 5 times with exponential backoff, simple and semantic caching, and 50+ prebuilt guardrails on inputs and outputs. You can run it yourself for $0, the same starting point as LiteLLM.
The difference is the hosted side. Portkey's Revoyant listing describes a vendor-run control plane for logs, prompt management, guardrails and budgets, with a hybrid option that keeps the data plane in your VPC. Palo Alto Networks acquired Portkey in 2026, and portkey.ai/pricing now redirects to a Palo Alto Networks contact form, so no hosted price is published (Nov 2026).
What you give up is a public price and an independent vendor. The GitHub README marks usage tracing and prompt template management as hosted or enterprise features, so the free gateway alone does less than LiteLLM's free proxy with its admin UI and budgets. Expect a security-vendor sales cycle for the managed product.
Choose Portkey over LiteLLM if you want guardrails and a dashboard run by a vendor, can accept a quote from Palo Alto Networks, and would rather not operate Postgres and Redis for the control plane yourself.
OpenRouter, the LiteLLM alternative for hosted routing
OpenRouter does LiteLLM's routing job as a service. You call one API, it forwards to the model provider, and token prices match each provider's list price on every plan. There's nothing to deploy, no database and no autoscaler.
The fee is on credits, not requests. Standard pay-as-you-go adds 5.5% to credit purchases and Business adds 8% (openrouter.ai, Nov 2026). Bring-your-own-key traffic is free up to $25,000 a month of list-price inference, then costs 5% of what the request would have cost on OpenRouter. The pricing page says it sets no plan-based rate limit on paid models, so limits come from the providers.
What you give up is self-hosting: every prompt passes through OpenRouter's servers, which some security reviews will rule out. The free plan is a sampler, capped at 20 requests a minute and 50 a day on free models, rising to 1,000 a day after you buy $10 of credits. The pricing page describes billing and limits, not LiteLLM-style budgets per internal team, so check OpenRouter's docs for key controls before you move several teams over.
Choose OpenRouter over LiteLLM if nobody on your team wants to own a gateway service and a 5.5% fee, or $0 under the BYOK allowance, costs less than the staff time.
Cloudflare AI Gateway, the LiteLLM alternative for low-cost logging
Cloudflare AI Gateway sits in front of your provider calls on Cloudflare's network. Its pricing page says dashboard analytics, caching and rate limiting are offered free (Nov 2026). For a team whose main use of LiteLLM is caching repeated prompts and capping request rates, that covers the job with no servers.
Logs are where the meter starts. Gateways created before 24 September 2026 keep 100,000 logs in total on Workers Free and 10,000,000 per gateway on Workers Paid. Newer gateways follow Workers Logs pricing: the Paid plan, $5 a month minimum, includes 20 million log events a month and charges $0.60 per extra million with 7-day retention; Free allows 200,000 events a day with 3-day retention (developers.cloudflare.com, Nov 2026). Cloudflare says Workers Logs moves to its Observability pricing on 1 December 2026, so recheck before you budget.
What you give up is LiteLLM's governance layer. The pricing page doesn't describe virtual keys, per-team budgets or spend tracking by tag, and you can't run the gateway in your own data center. Logpush export is billed separately by volume.
Choose Cloudflare AI Gateway over LiteLLM if you already use Cloudflare and want caching, rate limits and request logs for $0 to $5 a month rather than a self-hosted proxy.
Kong AI Gateway, the LiteLLM alternative for API platform teams
Kong AI Gateway puts LLM traffic under the same control plane as the rest of your APIs. For teams that already route REST and gRPC traffic through Kong, it means one set of policies, one analytics view and one vendor instead of a separate Python proxy.
Konnect Plus starts from $25 a month plus usage, billed monthly, with a 30-day trial and no card (konghq.com, Nov 2026). It includes 1 million AI gateway requests a month, charges $200 for each extra million and caps at 10 million. LLM model proxies cost $100 a month per model, up to 5 models, and analytics past the first million requests cost $20 per million. Support on Plus is email only with a 2-business-day SLA.
What you give up is LiteLLM's free price at volume. At 3 million requests a month across three models, Plus runs to roughly $765 a month before bandwidth. RBAC, SSO, audit logging and Kong's paid AI plugins sit in Enterprise, which is custom and annual, and fully self-hosted gateways are a separate Gateway Enterprise product priced by sales.
Choose Kong AI Gateway over LiteLLM if your API platform team already runs Kong and wants LLM routing inside the same policy and support contract.
Novita, the LiteLLM alternative for open-model access
Novita solves a different version of the problem. Many teams adopt LiteLLM to call several open-model hosts through one format. Novita hosts 200+ models behind one API itself, so if its catalogue covers your models, you get one key and one bill without running a gateway.
Prices are per million tokens. Novita lists Llama 3.3 70B Instruct at $0.135 input and $0.40 output, DeepSeek R1 0528 at $0.70 and $2.50, and Qwen3.8 Flash at $0.15 and $0.47 (novita.ai, Nov 2026). Batch inference carries a 50% introductory discount on supported models. The page says the platform is free to start but lists no free credits, and GPU hourly rates aren't on the pricing page.
What you give up is vendor neutrality. Novita serves its own catalogue, not OpenAI, Anthropic or Bedrock, so a team mixing closed and open models still needs a gateway in front. You also lose LiteLLM's per-team keys and budgets.
Choose Novita over LiteLLM if your real problem is the price of open-model tokens, not governing keys across several providers.
A year of LiteLLM vs its alternatives
Assume a platform team routing 3 million requests a month to three models, with $10,000 a month of model spend. Here's what each route adds on top of that spend for a year, at list prices. Model tokens themselves are excluded, since they're the same or set by the provider.
| Route | Added cost for 12 months | What it covers | What's not included |
|---|---|---|---|
| LiteLLM Open Source, self-hosted | $0 license | Proxy, admin UI, virtual keys, budgets, load balancing | Your Postgres, Redis and compute bill, plus staff time |
| LiteLLM Enterprise | Price on request | Adds SSO, SCIM, audit logs, support SLAs | Same infrastructure as Open Source |
| Portkey open-source gateway | $0 license | Routing, fallbacks, retries, caching, guardrails | Hosted control plane, quoted by Palo Alto Networks |
| OpenRouter Standard, buying credits | $6,600 (5.5% of $120,000 in credits) | Hosted routing, no servers | Self-hosting option |
| OpenRouter with your own provider keys | $0 (under the $25,000 a month BYOK allowance) | Hosted routing, no servers | Self-hosting option |
| Cloudflare AI Gateway on Workers Paid | $60 ($5 a month minimum) | Analytics, caching, rate limiting, 20 million log events a month | Per-team budgets; pricing changes 1 Dec 2026 |
| Kong Konnect Plus | About $9,180 (about $765 a month) | 1 serverless control plane, 3 model proxies, 3 million requests, analytics | SSO, RBAC and audit logs, which are Enterprise only |
| Novita | $0 platform fee | Hosted open models at Novita's token prices | Closed models and any gateway layer |
The cheapest rows are the two open-source gateways and Cloudflare, but only LiteLLM's free tier includes per-team budgets and spend tracking without a contract. OpenRouter flips from $6,600 to $0 if you bring your own keys, so the deciding question there is if you already hold provider contracts.
Kong is the only route whose cost climbs with request count. Every extra million requests adds $200 a month on Plus, so at the 10 million cap the request line alone reaches $1,800 a month.
What leaving LiteLLM costs you
LiteLLM speaks the OpenAI format, so most application code needs only a new base URL and key to point elsewhere. The work sits in what you've configured inside the proxy: virtual keys handed to teams, budgets, model aliases, fallback chains and logging callbacks to Langfuse, LangSmith, Arize Phoenix or OpenTelemetry.
Export your spend logs from Postgres before you shut the proxy down, since none of these tools imports LiteLLM's history. Then plan for each team to receive new keys. The biggest gap on each move:
- Portkey: the managed control plane has no published price and is now sold by Palo Alto Networks.
- OpenRouter: every prompt passes through a third party, and per-team budgets need checking in its docs.
- Cloudflare AI Gateway: no virtual keys or team budgets on the pricing page, and log pricing changes on 1 December 2026.
- Kong AI Gateway: SSO, RBAC and audit logs need the custom Enterprise plan.
- Novita: covers only its own model catalogue, so mixed closed and open setups still need a gateway.
When LiteLLM is still the right tool
Stay on LiteLLM if your platform team already runs Postgres and Kubernetes and needs keys, budgets and spend tracking per team across many providers. Nothing else here gives that governance layer for $0 in license fees, and LiteLLM's 140+ provider integrations cover Bedrock, Azure, Vertex AI and vLLM in one config.
Stay too if prompts can't leave your network. LiteLLM runs self-hosted or air-gapped, while OpenRouter and Cloudflare are hosted services and Kong's fully self-hosted option is a separate sales-led product. If SSO and audit logs are the only gap, run the 30-day Enterprise trial before you migrate. Browse more tools in AI infrastructure and LLMOps software.
Frequently asked questions
How much do LiteLLM alternatives cost?
It depends on who runs the gateway. Portkey's open-source gateway is $0 to self-host, Cloudflare AI Gateway's core features are free with Workers Paid at $5 a month, and Kong Konnect Plus starts from $25 a month plus usage. OpenRouter adds 5.5% to credit purchases on Standard (all checked Nov 2026).
What is the best open-source alternative to LiteLLM?
Portkey's gateway is the closest open-source alternative: MIT-licensed, with fallbacks, retries, caching and 50+ guardrails. Its free version lacks some of LiteLLM's free extras, since the README marks usage tracing and prompt management as hosted or enterprise features. LiteLLM itself remains the open-source option with the most governance built in.
Is OpenRouter a replacement for LiteLLM?
For routing, yes. OpenRouter gives one API for many models at provider list price, with a 5.5% fee on credits or no fee on your own keys up to $25,000 a month. It's hosted only, so it doesn't replace LiteLLM where prompts must stay in your network.
What is the difference between LiteLLM and Portkey?
Both are gateways with an open-source core. LiteLLM's free proxy includes an admin UI, virtual keys and budgets that you host on your own Postgres and Redis. Portkey pairs its MIT gateway with a vendor-run control plane, now sold by Palo Alto Networks with pricing on request.
Does LiteLLM have an enterprise version?
Yes. LiteLLM Enterprise adds SSO, SCIM, JWT auth, audit logs, secret-manager integration and support SLAs. The price isn't published on litellm.ai, but a 30-day self-hosted trial license is available without a sales call (Nov 2026).
Can Cloudflare AI Gateway replace a self-hosted LLM proxy?
It can replace the caching, rate limiting and logging parts. Cloudflare lists those features as free, with log volume billed through Workers Logs on newer gateways. Its pricing page doesn't describe virtual keys or per-team budgets, so teams that rely on LiteLLM for spend control should check the docs first.
Do I need an LLM gateway if I use one model provider?
Often not. If one host such as Novita serves every model you call, its single API and bill cover the routing job. A gateway earns its place once several providers, several teams or fallback chains are involved.
Sources
- LiteLLM plans (Open Source and Enterprise)
- LiteLLM docs: production best practices
- LiteLLM security update, March 2026
- Portkey pricing URL (redirects to Palo Alto Networks AI gateway contact form; no prices published)
- Portkey AI Gateway on GitHub (MIT license, features)
- OpenRouter pricing
- Cloudflare AI Gateway pricing
- Cloudflare Workers pricing (Workers Paid minimum, Workers Logs)
- Kong Konnect pricing (AI gateway)
- Novita model API pricing