GILD verdict
Pick LiteLLM when prompts and keys must stay in your network and you can staff a gateway, as Okta and Zalando do.
It charges nothing on model spend and caps it per person and team. OpenRouter is the better choice when you would rather pay 5.5% on credits than run, upgrade, and patch the gateway yourself: at 1 million Sonnet 5.5 requests a month that fee is about US$196 more than LiteLLM's VM, far less than an engineer's time. Teams that need both keep LiteLLM in front and route to OpenRouter behind it, with fallbacks to a direct provider.
Teams report reliability trouble on both sides. A developer moved from LiteLLM to OpenRouter over repeated outages (Mar 2026), while a DevOps engineer kept LiteLLM in front of OpenRouter and added direct OpenAI fallbacks after OpenRouter returned 429 errors in test production (Jul 2026). On GILD's workload of 1 million Claude Sonnet 5.5 requests a month, LiteLLM costs US$4,024 on one self-run VM and OpenRouter US$4,220 with its 5.5% fee on credits, so the real difference is the staff time LiteLLM needs. LiteLLM uses your provider keys inside your infrastructure; OpenRouter hosts the router and bills prepaid credits. Stripe agreed to acquire OpenRouter on August 19, 2026, and neither company had announced the close by October 28, 2026.
What teams report
Failures and incidents 1
FutureSearchOn LiteLLM
FutureSearch runs routing, fallback, retry and accounting through LiteLLM; a June 4, 2026 upgrade dropped the request tags its billing read, and usage went unbilled for 6.5 days (Jun 10, 2026).
“A routine LiteLLM version bump turned off our usage billing for six and a half days. We had many alerts for runaway spend, but not for spend dropping to zero.”
Jack Wildman, Senior Software Engineer, FutureSearchFutureSearch blog, , write-up
Left or turned down 1
SRVpressOn OpenRouter
The SRVpress developer stopped using OpenRouter over slowness and errors (X, Oct 25, 2026).
“Stopped using OpenRouter for same reason, slowness & errors.”
Jay Brunet, System Admin and Developer, SRVpressX, , post
Switched 2
Hacker News user (organization not stated)
The developer moved from LiteLLM to OpenRouter about a year before March 2026 because LiteLLM kept causing outages, and called the move lucky on the day of the PyPI compromise (Hacker News, Mar 24, 2026).
“Now I feel lucky that I switched to just using OpenRouter a year ago because LiteLLM was incredible flaky and kept causing outages.”
chillfox, Developer, Hacker News user (organization not stated)Hacker News, , post kisamoto, DeveloperOn OpenRouter
The author split a US$100 monthly Claude subscription budget into a US$10 Zed plan, a kept US$20 Cursor plan and US$70 a month of OpenRouter credits on ZDR-only endpoints, then edited the post to note OpenRouter's 5.5% fee (Apr 6, 2026).
“OpenRouter does charge a 5.5% fee.”
kisamoto, Developer, braw.dev (personal blog)braw.dev, , write-up
Adopted 1
ZalandoOn LiteLLM
Zalando's ML platform team has run a LiteLLM proxy since January 2024 as the company gateway to OpenAI, AWS Bedrock and Google Vertex models (Aug 14, 2026).
“…our ML platform team deployed in January 2024 a LiteLLM based API proxy with access to models from different providers (now: OpenAI, AWS Bedrock, and Google Vertex).”
Bartosz Ocytko, Executive Principal Engineer, ZalandoZalando Engineering Blog, , write-up
Choosing between LiteLLM and OpenRouter
Choose LiteLLM if
- Prompts and provider keys must stay in your own network, or your contracts require calling providers directly.
- You want budgets and rate limits per person, team, and app without paying a fee on model spend.
- You run self-hosted models, such as vLLM or Ollama, beside hosted providers behind the same endpoint.
- Your services already call one gateway and you want hosted providers added behind it with fallbacks set centrally, as an RTFM engineer did with OpenRouter in July 2026.
Choose OpenRouter if
- Nobody on the team should operate, upgrade, or patch a gateway; a developer who moved from LiteLLM to OpenRouter in 2025 did so over outages.
- You want 500+ models from 80+ providers without setting up an account with each provider.
- You want fallback across providers where failed attempts are not billed, and EU or US in-region routing on the Business plan.
Side by side
| Dimension | LiteLLM | OpenRouter |
|---|---|---|
| What you payList prices in US dollars. LiteLLM adds your own infrastructure cost; OpenRouter does not mark up tokens. | Open source US$0 to self-host, with no per-token fee; Enterprise is an annual license quoted by gateway request capacity, deployment architecture, and support needs, never per token | Provider list price for tokens, plus a fee on credit purchases: 5.5% on Standard, 8% on Business (a tier added in September 2026), discounted on Enterprise contracts |
| Monthly bill at one workloadGILD's calculation from published rates, fetched Oct 28, 2026: 1 million requests a month, each with 1,000 input and 200 output tokens on Claude Sonnet 5.5 at US$2 and US$10 per million tokens, no caching. The LiteLLM line has no high availability, Enterprise license, support or engineer time; the OpenRouter line assumes its Sonnet route lists Anthropic's rate. Taxes are excluded. | US$4,024 a month: US$4,000 of Claude Sonnet 5.5 tokens at Anthropic's list price plus US$24 for one 2-vCPU, 4-GiB VM running the gateway, Postgres and Redis11828 | US$4,220 a month on Standard credits: the same US$4,000 of tokens plus the 5.5% fee on the credit purchase; US$4,000 with your own provider keys, inside the US$25,000 a month of BYOK spend that carries no fee81328 |
| Reliability teams report | A developer left after repeated outages (Mar 2026); a June 2026 upgrade stopped FutureSearch's usage billing for 6.5 days without an error; Zalando restarts its proxy every 20,000 requests over memory leaks (Aug 2026)202122 | An RTFM engineer hit 429 errors through OpenRouter in test production and added direct OpenAI fallbacks in LiteLLM (Jul 2026); an SRVpress developer stopped using OpenRouter over slowness and errors (Oct 25, 2026)2425 |
| Who runs it | Self-hosted only in your cloud or data center, from Docker images, a Helm chart, or Terraform modules; air-gapped deployments on Enterprise | Hosted by OpenRouter; no self-hosted option |
| Your own provider keys | Required: you bring your own keys, and LiteLLM calls each provider with your keys and contracts19 | Optional BYOK: US$25,000 of list-price inference a month with no fee on Standard and Business, then a 5% fee13 |
| Models | 140+ providers and 1,892 models listed on BerriAI's site, called with your own provider keys and contracts | 500+ models from 80+ providers on paid plans; 25+ free models with 50 requests a day on the free plan |
| Spend controls | Budgets and RPM and TPM limits per key, user, team, and organization in open source, enforced against a Postgres database; tag budgets, soft budget alerts, and temporary increases need Enterprise | Per-key credit limits with daily, weekly, or monthly resets and workspace budget alerts, from the Standard plan |
| Prompt data | Stays in your infrastructure; self-hosted LiteLLM keeps prompts and telemetry away from BerriAI | Not stored unless you opt in to logging or to OpenRouter's use of your data, or content is flagged as potential CSAM; request metadata is stored; ZDR-only routing from the Standard plan, since the pricing table lists data-policy routing as unavailable on Free; on October 27, 2026 OpenRouter announced it will screen all traffic, ZDR included, for CSAM; EU and US in-region routing on Business and Enterprise |
| Single sign-on | SSO free for up to 5 users from v1.76.0; beyond that, and SCIM, audit logs, and secret managers, an Enterprise license | SSO (SAML) and SCIM on Enterprise; Workload Identity Federation on Business and Enterprise |
| Failover | Load balancing, retries, fallbacks, and cooldowns across deployments you configure | Routing and fallback across providers; attempts that error are not charged for model tokens |
| Security record | 44 reviewed GitHub advisories for the package as of October 28, 2026. Two PyPI releases (1.82.7 and 1.82.8) carried credential-stealing malware on March 24, 2026, and 1.107.0 on PyPI and four official Docker tags carried it on October 21, 2026. SOC 2 Type II audited | A hosted service; its Enterprise page states SOC 2 compliance and GDPR compatibility |
| Ownership | BerriAI, a Y Combinator W23 company17 | OpenRouter; Stripe agreed to acquire it on August 19, 2026, terms undisclosed, and neither company had announced the close as of October 28, 202616 |
What practitioners report, side by side
Each product's figures come from its own practitioners, for the group you pick.
Overall signal
LiteLLM reads +10 and OpenRouter +10, each from its own practitioners. LiteLLM: 38 positive, 28 negative of 108. OpenRouter: 41 positive, 30 negative of 101.
The score counts the authors from all practitioners that take a side, one vote each with their latest view, and is how far positive votes lead negative ones, per 100 votes, so LiteLLM's +10 rests on votes split 35 positive, 42 mixed and 25 negative and OpenRouter's +10 on 40, 30 and 30.
A lead of a fifth of the votes reads positive and a lead of half strong positive, with the same lines below zero. The dots show each source, positive, mixed, negative. Run a pilot on one service's model traffic to measure added latency, failed fallbacks and how closely its spend records match each provider's bill.
Footprint on the AI Application Platform stack
They differ on two of 9 layers. On Model APIs, LiteLLM touches it and OpenRouter supports it. On Tool use, LiteLLM supports it and OpenRouter touches it. Coverage is each product's placement on this stack, as its product page shows it. Core is a job the product owns; Supports and Touches are jobs it helps with. Coverage says nothing about how well a product does them.
| Layer | LiteLLM | OpenRouter |
|---|---|---|
| Model | ||
| Model APIs | Touches | Supports |
| Build | ||
| Tool use | Supports | Touches |
The 7 layers where they match
| Layer | LiteLLM | OpenRouter |
|---|---|---|
| Model | ||
| Model gateway | Core | Core |
| Build | ||
| Orchestration | Touches | Touches |
| Data | ||
| Retrieval | Touches | Touches |
| Evaluate | ||
| Evaluation | Touches | Touches |
| Operate | ||
| Observability | Supports | Supports |
| Cloud deployment | Touches | Touches |
| Guardrails | Supports | Supports |
What practitioners praise
Among all practitioners, LiteLLM alone has five praise themes that reach 5 sources from 3 authors. Virtual Keys, Budgets & Spend Tracking and Self-Hosted Control lead, at 19 of 24 and 12 of 12. OpenRouter alone has three praise themes that reach 5 sources from 3 authors. One API and One Balance and Fast Access to New and Open Models lead, at 25 of 25 each.
| Praised for | LiteLLM | OpenRouter |
|---|---|---|
| Virtual Keys, Budgets & Spend Tracking | 19 of 24 | Not a theme in its sources |
| Self-Hosted Control | 12 of 12 | Not a theme in its sources |
| One API Across Providers | 12 of 24 | Not a theme in its sources |
| Routing, Fallbacks & Load Balancing | 8 of 10 | Not a theme in its sources |
| Provider & Model Coverage | 2 of 7 | Not a theme in its sources |
| One API and One Balance | Not a theme in its sources | 25 of 25 |
| Fast Access to New and Open Models | Not a theme in its sources | 25 of 25 |
| Spend Limits & Cost Data | Not a theme in its sources | 13 of 16 |
Each count is the sources that praise a theme, or raise it as a concern, out of those that discuss it. A row pairs the two products when both carry a theme under the same or an equivalent name.
A count ending in "so far" has under 5 sources from 3 authors among all practitioners. "Not a theme in its sources" means GILD found no theme of that name in that product's sources.
What practitioners raise as concerns
Among all practitioners, LiteLLM alone has five concern themes that reach 5 sources from 3 authors. Stability & Code Quality and Performance at Scale lead, at 29 of 29 and 14 of 16. OpenRouter alone has four concern themes that reach 5 sources from 3 authors. Fees & Billing Surprises and Provider Quality Variance lead, at 20 of 25 and 16 of 18.
| Raised as a concern | LiteLLM | OpenRouter |
|---|---|---|
| Stability & Code Quality | 29 of 29 | Not a theme in its sources |
| Performance at Scale | 14 of 16 | Not a theme in its sources |
| Security & Supply Chain | 10 of 14 | Not a theme in its sources |
| Operational Burden | 9 of 10 | Not a theme in its sources |
| Enterprise-Gated Features | 6 of 6 | Not a theme in its sources |
| Fees & Billing Surprises | Not a theme in its sources | 20 of 25 |
| Provider Quality Variance | Not a theme in its sources | 16 of 18 |
| Latency, Errors & Rate Limits | Not a theme in its sources | 14 of 16 |
| Free-Model Limits | Not a theme in its sources | 3 of 5 |
28 sources
- LiteLLM pricing
- LiteLLM docs: Enterprise
- LiteLLM docs: Budgets and rate limits
- LiteLLM docs: Data privacy and security
- LiteLLM docs: SSO for the Admin UI
- LiteLLM docs: Load balancing
- LiteLLM AI Gateway home page (providers, models, adoption figures)
- OpenRouter pricing and plan comparison
- OpenRouter Enterprise
- OpenRouter docs: Data collection
- OpenRouter docs: Provider logging and in-region routing
- OpenRouter docs: Zero Data Retention
- OpenRouter docs: Bring your own keys
- OpenRouter blog: CSAM Screening Across OpenRouter (Oct 27, 2026)
- OpenRouter blog: OpenRouter is Joining Stripe (Aug 19, 2026)
- Stripe newsroom: Stripe agrees to acquire OpenRouter (Aug 19, 2026)
- BerriAI/litellm on GitHub (stars, forks, releases)
- LiteLLM docs: Production best practices
- LiteLLM docs: Providers
- Hacker News comment on switching from LiteLLM to OpenRouter (Mar 24, 2026)
- Jack Wildman, FutureSearch blog: LiteLLM made us accidentally make our product free for a week (Jun 10, 2026)
- Bartosz Ocytko, Zalando Engineering Blog: Agentic engineering at Zalando, a snapshot (Aug 14, 2026)
- Dennis Henry, Okta blog: How Okta deploys and secures AI for its employees (Feb 26, 2026)
- Arseny Zinchenko, RTFM: LiteLLM, OpenRouter integration and fallbacks configuration (Jul 31, 2026)
- X post by SRVpress on leaving OpenRouter over slowness and errors (Oct 25, 2026)
- braw.dev: Reallocating US$100 a month of Claude Code spend to Zed and OpenRouter (Apr 6, 2026)
- FutureSearch: company page listing Jack Wildman as Senior Software Engineer and Callum McMahon as Research Scientist (Oct 30, 2026)
- Anthropic: Claude API pricing, Claude Sonnet 5.5 at US$2 and US$10 per million tokens (Oct 28, 2026)