LiteLLM alternatives: 6 tools compared (October 2026)

Shortlist: which LiteLLM alternative fits which situation

Portkey (Prisma AIRS AI Gateway)

A match for teams that want a hosted gateway with logs, guardrails and prompt management and accept that Palo Alto Networks now owns the vendor. Production lists at $49 a month for 100,000 recorded logs, and the open-source gateway is MIT licensed.

OpenRouter

Suits teams that want one API and one bill for 500+ models without running servers. Standard adds 5.5 percent on credit purchases, and with your own provider keys the first $25,000 of list-price usage a month carries no fee.

Orq.ai

Fits teams that need an EU-built, EU-hosted gateway with an OpenAI-compatible endpoint, per-team budgets and PII redaction. Pay-as-you-go has no base fee and a 4.5 percent fee on Orq.ai credits.

Dataiku

Only worth a look if you already run Dataiku for data and ML work. Its LLM Mesh puts quotas and cost controls on model access inside the platform, and the vendor publishes no price.

What a gateway costs at 100,000 and 1,000,000 requests a month

The vendors do not bill the same thing. LiteLLM's open-source code costs nothing to license and moves the bill to your servers. Portkey bills recorded logs, OpenRouter and Orq.ai bill a percentage of credits, and Dataiku sells a platform. So I fixed one reference: 100,000 requests with $1,000 of model spend a month, then ten times that, and applied each vendor's listed rule.

Running LiteLLM yourself is not free of effort. The production guide asks for a pod of 1 vCPU and 4 Gi of memory as requests and limits, PostgreSQL for virtual keys, and Redis 7.0 or newer once more than one instance runs. If the database goes down, keys already in the cache keep working for 60 seconds by default and uncached key lookups fail. The vendor publishes no server cost, so I cannot give you a number for that part.

What this means for you: at these volumes a hosted fee is between $49 and $800 a month, which is often less than the time of one engineer who patches and monitors a proxy. It is more than $0, though, and it grows with spend for the percentage vendors. Put your own request count and spend into the table before you decide.

Sources: [1][2][3][4][5][6][7]

Stay or switch: license, features and who runs it

Not everyone should leave. LiteLLM's repository had 61,283 stars when I pulled the GitHub record on October 15, 2026, and its latest release, v1.105.1, came out on October 14. The project is active, and its LICENSE file puts everything outside the enterprise/ directory under MIT.

License. The enterprise/ directory has its own license. It lets you copy and modify the software for development and testing without a subscription, and it forbids production use without a valid subscription for the right number of seats. Teams that switched on SSO or audit logs from that directory during a trial and kept it running should treat this as the first reason to act.

Features. Two LiteLLM pages disagree. The Enterprise page lists virtual keys with budgets under its governance features, while the home page lists virtual keys, budgets and teams in the free tier, and the key documentation I read marks key rotation as an Enterprise feature. SSO, SCIM, JWT authentication, audit logs and RBAC are Enterprise on both vendor pages. Test your exact feature set against the 30-day trial key before you pay or leave.

Operations. A gateway holds every provider key, and LiteLLM's own security notices show what that costs to run. The box below lists them with dates. The vendor names the affected versions, which is a point in its favor.

What this means for you: if nobody owns patching and version pinning, a hosted gateway moves that job to a vendor, and the price table shows the cost. If you must keep prompts and logs on your own servers, leaving may cost more than it saves.

Sources: [8][9][2][10][11][12][13][14][15][16]

The four alternatives, and who each one is not for

Portkey sells a hosted gateway with logs, guardrails, prompt management and role-based access. Developer is free with 10,000 recorded logs and 3 days of log retention, and Portkey marks it not suitable for production. Production is $49 a month, keeps logs 30 days and metrics 90 days, and its overage rule runs up to 3 million logs. SSO, a private cloud and data export sit in Enterprise. Portkey's website now reads Prisma AIRS AI Gateway after the acquisition, and the press release promises continued support for customers without mentioning the open-source gateway. That repository is MIT licensed, and its last push was on May 25, 2026. If you chose Portkey for its independence, that changed on May 29, 2026.

OpenRouter is the opposite of LiteLLM. You run nothing, and one API reaches 500+ models from 80+ providers. The FAQ says it passes provider prices through without markup on inference, and the fee is on credit purchases. By default it does not store prompts or responses, and logging is opt-in. EU and US routing exist only on Business and Enterprise, and it fails closed if no provider in the region serves the model. It is not for teams that need per-team spend controls inside their own network, or SSO on the Standard plan.

Orq.ai is built in Amsterdam and hosted in the EU by default. Its gateway offers an OpenAI-compatible endpoint to 500+ models from 30+ providers, budgets per user and team, and PII redaction. The pricing page lists SOC 2 Type II and ISO 27001. Orq.ai deletes observability data after 14 days on Free and 30 days on pay-as-you-go. Teams that want to keep every trace for a year need a custom plan, and teams that only want a proxy will pay for more platform than they use.

Dataiku is a data and ML platform, and its LLM Mesh is a gateway layer inside it. The product page lists quotas, budget enforcement and usage tracking, and the documentation index shows an OpenAI-compatible API section. I read only the index, not the details. It is for companies that already pay for Dataiku. A team that just wants to replace a proxy should look at the other three.

What this means for you: a hosted control plane points to Portkey, one bill and no servers to OpenRouter, EU hosting by default to Orq.ai, and an existing data platform to Dataiku.

Sources: [3][17][18][16][19][4][5][20][21][6][22][23][7][24]

What leaving costs: terms, cancellation, export, deletion

Because LiteLLM is software you run, leaving it is mostly configuration. The model list lives in config.yaml, optionally also in the database. Spend is stored per key, user and team in a SpendLogs table, and the /spend/logs/v2 endpoint pages through up to 1,000 rows at a time. Clients that already speak the OpenAI format change a base URL and a model name.

The harder part is the contract with the next vendor. I read each one's terms on October 15, 2026. OpenRouter lets you terminate any time but refunds unused credits only within 24 hours, expires them after 365 days, and does not refund platform fees. Orq.ai ends a subscription at the end of the term without partial refunds and asks for two months of written notice to switch providers. Portkey's terms page contains no contract text, so I cannot state its term or notice period. Dataiku's addendum names deletion after termination without a deadline.

What this means for you: the notice period is rarely the expensive part. Paying two gateways during the parallel run, re-issuing keys and rebuilding fallbacks cost more. Keep credits on pay-as-you-go vendors low so expiry hurts little, and ask Portkey for its contract before you sign.

Sources: [25][26][27][9][28][29][30][31][32][33][34][35][36]

Corrections, disagreements and what I left out

I updated one value from our own data. Our LiteLLM profile quoted about 59,800 GitHub stars from September 28; the GitHub record on October 15 shows 61,283. The other entries I checked, such as Portkey's $49 plan, OpenRouter's 5.5 percent fee and Orq.ai's 4.5 percent fee, matched the vendors' pages.

Sources disagreed in two places. LiteLLM's documentation says 100+ LLMs while its home page says 140+ providers, and its Enterprise page and home page place virtual keys in different tiers. Portkey's repository shows 13,191 stars through the GitHub record, while its site footer says 10.2K.

I left out Msty, a desktop client for language models, and Artificial Analysis, a benchmark site. Neither routes production traffic. I also left out review-site opinions and Reddit threads, because they are not primary sources, and I did not cost Portkey's Enterprise plan, which has no public price.

Sources: [12][27][10][2][19][3][4][6]