Compare up to four

Every value links to the page it came from, so you can check any figure in one click. Blank cells read "Not published" rather than "No" — we do not treat silence as a negative answer. Turn on Technical in the header for the finer fields and precise terminology.

OpenRouterVercel AI Gateway

Only show differences

Feature and pricing comparison of OpenRouter, Vercel AI Gateway
Attribute	OpenRouterManaged marketplace	Vercel AI GatewayManaged gateway
Overview
What kind of productMarketplaces resell many providers behind one key. Gateways add governance on top. Open-source projects you run yourself. Cloud platforms are hyperscaler surfaces. Inference providers host models on their own hardware.	Managed marketplace	Managed gateway
Who runs itManaged means the vendor operates it. Self-host means you run it on your own infrastructure. Both means you can choose.	Managed only	Managed only
LicenceProprietary products cannot be inspected or forked. Open licences such as MIT and Apache-2.0 let you audit, modify, and run the code without permission.	Proprietary	Proprietary
CompanyThe organisation that maintains the product.	OpenRouter, Inc.	Vercel
Who you would be signing withWhether the product is still an independent company, has been acquired, is a large cloud vendor’s product line, is run by a software foundation, or has been put into maintenance mode. Maintenance mode means bug fixes and security patches only — no new features.	Being acquired — deal not yet closed	Independent company
Last shipped an updateThe date of the most recent release or version tag. A product that has not shipped in a year is a different risk from one that shipped last week, regardless of what its marketing site says.	2026-08-19	2026-08-28
Cost
Markup on model pricesHow much the product adds on top of what the underlying model provider charges. Zero means you pay the same per-token price you would pay the model provider directly.	None	None
Fee to add fundsA percentage charged when you top up your balance, separate from token prices. It is easy to miss because it does not appear on the per-token price list.	5.5%	Nonebest
Monthly cost per personA recurring per-user platform charge that applies regardless of how much you use the models.	Not published	None
Can use your own provider accountsBring Your Own Key: you keep direct contracts with OpenAI, Anthropic and others, and the gateway only routes traffic. This preserves negotiated rates and committed-spend discounts.	Yes	Yes
Cost of using your own accountsWhat the product charges to route traffic through your own provider keys.	Free up to $25,000/mo of list-price inference (pay-as-you-go) or $200,000/mo (Enterprise), then 5% of the equivalent OpenRouter cost; measured by list-price cost, not request count.	No markup or fee on BYOK, but BYOK requires purchased AI Gateway Credits (paid tier); failed BYOK requests fall back to Vercel system credentials and are billed against credits, and BYOK spend is not covered by budgets.
Free tierWhat you can do without paying, useful for evaluation.	Free tier with 25+ free models and 50 free-model requests/day; 1,000/day after purchasing at least $10 in credits.	Every team gets free AI Gateway credits on a subset of models with lower per-model rate limits; the amount is not published. Pro plans include $20 of credit across resources.
Enterprise plan fromAnnual entry price for the enterprise tier, where one is published or credibly reported.	Not published	Not published
How the vendor makes moneyThe shape of the vendor’s bill: does the routing layer charge a percentage on top of tokens, a flat monthly fee, both, neither (because inference is the product), or nothing at all (open source with no paid tier).	Percentage on tokens or top-ups	Percentage on tokens or top-ups
How pricing works, brieflyA one-paragraph description that covers the caveats a pricing category cannot: introductory rates, per-feature meters, tier gating, and pricing that resets on a specific date.	Credit top-up fee (5.5% Stripe / 5% crypto) with a $0.80 minimum; 0% token markup; BYOK charged 5% above a monthly list-price allowance ($25k Pay-as-you-go / $200k Enterprise).	Gateway credits are pay-as-you-go with no token markup or mandatory platform subscription. Optional governance, reporting and trace exports have separate charges; some require Pro or Enterprise. Include only incremental platform costs for the selected features.
Minimum commitmentWhether the vendor requires a minimum contract term, a minimum spend, or a provisioned-capacity purchase to get its published rate.	None stated. Unused credits may expire one year after purchase.	None. Credits purchasable at any time with no obligation to renew; custom volume discounts on Enterprise.
Charges that fire after you go over an allowanceThe line items that scale with usage after an included allowance is exhausted — log storage, extra requests, per-feature meters, data export — which is where cost estimates usually go wrong.	No log/trace-retention or request-volume overage. Only after-the-fact charge is the BYOK 5% once the monthly list-price allowance is exceeded.	Trace Drains meter trace events delivered and trace data transferred; Pro plans include no allowance for either. Team-wide provider allowlist: $0.10 per 1,000 successful requests. Team-wide zero data retention: $0.10 per 1,000 requests. Custom Reporting: $0.075 per 1,000 tag/user/quota-entity writes and $5 per 1,000 reporting-endpoint queries. All billed outside credits.
Prompt cache offeredWhether the gateway offers its own response cache, what kind of match it does (exact request, prefix, semantic), or simply passes provider caching through unchanged.	Passes provider caching through	No gateway-owned cache
Discount on cached inputHow much cheaper cached tokens are than fresh input, when the vendor publishes a single figure. Bundled-inference clouds usually price this per model instead of as one number.	Not published	Not published
Premium on cache writesHow much more the first write of a cached prefix costs versus a plain input token. A high write premium and a low hit rate can leave you paying more than you save, so this matters as much as the read discount.	Not published	Not published
Who captures the cache savingWhether the customer keeps the full saving from caching or the vendor captures part of it — and any conditions attached (write premium, storage fees, best-effort hits).	Provider caching flows through unchanged. Read discounts vary by provider: 75%/50% for OpenAI, 90% for Anthropic/Alibaba/DeepSeek, 50% for Groq, ~80% for Z.AI, 75% for Gemini implicit. Cache-write premium: 0% for pre-GPT-5.6 OpenAI/Grok/Moonshot/Groq/Gemini; +25% for GPT-5.6+, Alibaba explicit and Anthropic 5-min; +100% for Anthropic 1-hour. OpenRouter itself adds 0% on cached traffic.	Pricing page does not state whether the gateway caches or passes through provider caching, and no cached-token pricing is published. Because Vercel charges provider list price with 0% markup, any provider cache discount would reach the customer unchanged.
What you can split spend byThe dimensions the vendor documents for splitting spend — per key, per user, per team, per tag, per customer. Matters if you need to chargeback internally or bill an end customer.	Per model, provider and API key on the Activity page; session_id grouping across turns. Per-user/team/customer not stated.	By tags, user IDs and quota entity IDs via Custom Reporting (queried through the reporting endpoint at $5/1k queries). Per team/key/customer not explicitly stated.
How you get cost data outThe mechanisms the vendor publishes for exporting cost and usage data: CSV, an API, webhooks, S3, a data warehouse, or nothing at all. Any per-unit price is included.	API only (/api/v1/key, credits API, /api/v1/generation). CSV, webhook, S3 and warehouse export not stated.	Reporting API plus Trace Drains (Pro/Enterprise, metered). CSV/S3/warehouse not stated.
Who pays the model billWhether you bring your own provider accounts (BYOK), buy inference from this vendor, or can do either. This is the single biggest commercial difference between these products: it decides who holds the contract with the model provider and who carries the spend.	Your keys or their credits	Your keys or their credits
Catalog
Models availableHow many models you can call, shown as a range because several vendors publish different totals on different pages. Vendor-reported either way, so counts are not directly comparable — some count every provider variant of the same model separately.	458	386
Model providers reachableHow many distinct model providers or labs you can reach, shown as a range where the vendor’s own pages disagree. More providers usually means better redundancy when one has an outage.	~83	Not published
Works with standard OpenAI codeIf yes, you can usually switch to it by changing one base URL, and switch away just as easily. This is the main defence against lock-in.	Yes	Yes
OpenAI chat endpointThe endpoint almost every application ports first. "Not documented" means the vendor never states it, which is different from a documented no.	Yes	Yes
Anthropic messages endpointWhether Anthropic-shaped calls work without rewriting them. Several products support this only as SDK compatibility or provider passthrough rather than a native endpoint — the detail page says which.	Yes	Yes
OpenAI Responses endpointThe newer stateful OpenAI surface. Support is much thinner across this market than chat completions.	Yes	Yes
Embeddings endpointWhether you can generate vectors through the same gateway, or need a second integration for retrieval workloads.	Yes	Yes
Image generation endpointWhether image models are reachable through the same surface as text.	Yes	Yes
Audio endpointsSpeech-to-text and text-to-speech. Frequently the first gap in an otherwise complete gateway.	Yes	Yes
Batch jobs endpointAsynchronous bulk processing, usually at a discount. Commonly undocumented, and commonly the reason a migration stalls late.	Yes	Not published
Needs the vendor’s own code libraryA proprietary client library spreads through your codebase and has to be torn out again if you leave. “No” is the better answer here, and it means the standard OpenAI client works.	No	No
You can export your request historyWhether you can get your own request logs, traces, or usage records back out — through an API, a bulk export, or a download. Decides whether you leave with your history or abandon it.	Yes	Yes
Routing & reliability
Uptime it promises in writingThe uptime percentage in a published, contractual service level agreement. A public status page is not an SLA — it reports what happened, it does not promise anything or pay you back when it breaks.	Not published	99.99%
Automatic failoverWhen a model provider goes down or rate-limits you, traffic moves to a backup automatically instead of returning errors to your users.	Yes	Yes
Load balancingSpreads requests across several providers or keys to raise your effective rate limit.	Yes	Yes
Rule-based routingSend different requests to different models based on rules — for example a cheap model for free users and a strong model for paying ones.	Yes	Yes
Response cachingReuses the answer when the exact same request comes in again, which cuts both cost and latency.	Yes	Yes
Similar-question cachingReuses an answer when a new question means roughly the same thing as an earlier one. Saves far more than exact-match caching but can return subtly wrong answers if tuned loosely.	No	No
Where you set the timeoutWhere a request timeout can be set: per request, in a config file, in the vendor dashboard, or nowhere. Nine of the twenty products documented here do not describe a request timeout at all, so the worst case of a hung upstream call is unknowable from the docs.	Not published	Per request
Where you set retriesWhere retry count and backoff are configured. Worth knowing alongside billing: a retried streaming call can be charged more than once.	Not published	Not published
Where you set fallbacksWhere the fallback chain is defined. Almost every product claims fallback; the useful question is whether you can change it from code or only by hand in a dashboard.	Per request	Per request
Upstream health trackingWhether the product notices a failing upstream and stops sending traffic to it, and whether you can tune the thresholds. This is what turns a provider outage into a blip rather than a sustained error rate, and only four of the twenty expose it.	Fixed, cannot change	Fixed, cannot change
Cross-region failover you controlWhether you can define what happens when a region degrades. A vendor running many regions is not the same as a vendor letting you configure failover between them; only three document a user-controlled mechanism.	Not published	Not published
Where you set load balancingWhere traffic distribution across upstreams or keys is configured.	Per request	Per request
Operations
Usage dashboards and logsBuilt-in visibility into what was sent, what came back, what it cost, and how long it took.	Yes	Yes
Spending limitsHard caps that stop spend before it becomes a surprise invoice. The single most valuable control for a small team.	Yes	Yes
Rate limitsCaps on request volume per key or per user, useful for protecting against abuse and runaway loops.	Yes	Yes
Separate keys per team or appIssue scoped keys with their own budgets and permissions so you can attribute cost and revoke access without rotating everything.	Yes	Yes
What gets loggedWhether full prompts and responses are stored, only metadata, or nothing. Full-body logging is the most useful debugging feature here and the one most likely to need a conversation with your compliance team.	Metadata only	Your choice
You can turn logging offWhether prompt and response bodies can be suppressed while still keeping usage metrics. Nineteen of the twenty document a way to do this; the mechanisms range from a per-request header to an organisation-wide setting.	Yes	Yes
Traces you can take elsewhereWhether the product emits OpenTelemetry, a proprietary format, or nothing. OpenTelemetry means the traces land in the tooling you already run instead of only in the vendor’s dashboard.	Not published	OpenTelemetry
Where telemetry can goDocumented sinks for logs and metrics. This is a good proxy for how replaceable the vendor’s own dashboard is: around twenty destinations means you never have to depend on it, while a CSV download means you do.	CSV export	CSV export
Performance
Delay it addsExtra time the product itself adds to each request, on top of however long the model takes. Usually irrelevant next to multi-second model latency, but it matters for high-volume or streaming-sensitive workloads.	Not published	Not published
What the request path runs onThe comparable way to talk about latency here. An edge worker, a compiled Go or Rust binary, and a Python proxy have different overhead floors no matter which figures each vendor publishes. Products that never disclose their runtime are recorded as undisclosed rather than assumed.	Edge worker	Undisclosed vendor service
You can run the request path yourselfWhether the component that actually carries your prompts can run on your own infrastructure. Distinct from a vendor offering a self-hosted control plane while still proxying traffic through their network.	Not published	No
Streaming responsesWhether token-by-token streaming is documented. The caveats matter more than the yes: some products cannot cancel a stream without still being billed, and several timeout and fallback mechanisms stop applying once the first token has been sent.	Yes	Yes
Security & compliance
Does your prompt reach their serversWhether the text you send passes through this company’s own infrastructure. If it does, every other promise on this page is a policy commitment rather than a physical impossibility. Self-hosted products can answer no outright.	Yes	Yes
What they keep if you change nothingDefaults matter more than options. A product that stores full prompts and replies unless you find the right header will have stored them by the time you read the docs.	Metadata only, not content	Yes — prompts and replies
How long they keep itDefault retention for request content, in days. Zero means nothing is kept. Read the note: several products keep nothing as a rule but make timed exceptions for abuse review or specific models.	0	30
Could they train on your promptsWhether the vendor may use your prompts and outputs to train models. “Not published” means we could not find any position, which is not the same as a no — ask for it in writing.	No	No
Where it runs, and what you can pinWhich regions are offered and whether you can force processing to stay in one. A global endpoint that silently picks a region is a different compliance story from an endpoint you pin yourself.	US servers by default. Enterprise adds EU region locking and in-region routing. No self-hosted or customer-VPC option.	US primary, and the data-processing agreement permits transfer anywhere in the world. No customer-VPC option and no documented gateway region pinning.
Where safety filters runA filter that strips personal data only helps if it runs before the data leaves your boundary. If guardrails execute in the vendor’s cloud, the vendor has already received whatever you wanted redacted.	In the vendor’s cloud	No guardrails offered
Who else touches the dataThe published list of third parties the vendor passes your data to. No list means you cannot know the full chain, which most data-protection agreements require you to.	https://openrouter.ai/authorized-sub-processors	https://security.vercel.com/
SOC 2 auditedAn independent audit of security controls. Enterprise buyers and their procurement teams routinely require it.	Yes	Yes
Will sign a HIPAA agreementRequired before you may send protected health information through the service. Without a signed BAA, healthcare data is off limits.	Not published	Yes
GDPR commitmentsPublished data processing terms for handling personal data of people in the EU and UK.	Yes	Yes
Can keep data in the EURequests can be processed inside the EU rather than routed to US infrastructure. Often the deciding constraint for European customers.	Yes	Not published
Does not retain your dataPrompts and responses are not stored after the request completes. Sometimes a paid add-on rather than the default.	Yes	Yes
Strips personal dataDetects and removes identifiers such as names, emails, and card numbers before the request reaches the model provider.	No	Not published
Content guardrailsPolicy checks on inputs and outputs — blocking unsafe content, enforcing formats, or catching prompt-injection attempts.	No	No
Blocks personal data in promptsWhether personal data detection sits on the request path and can stop the call, merely inspects and forwards it, or is not documented. A control that only reports is a logging feature, not a policy control.	Can block the request	Not published
Blocks prompt injectionWhether injection and jailbreak detection can stop a request. Most products offering this call a partner classifier rather than shipping their own.	Can block the request	Not published
Blocks harmful contentWhether hate, violence, sexual and self-harm categories are checked inline and can stop a request, in either direction.	Not published	Not published
Your own policy rulesWhether you can add your own rule — a regex, a webhook, or your own classifier — rather than choosing from the vendor library.	Can block the request	Not published
Where guardrails runWhether guardrail evaluation happens inside your infrastructure or on the vendor’s servers. This decides whether the prompt you are trying to protect leaves your network in order to be checked.	On the vendor's servers	On the vendor's servers
If the guardrail itself failsWhat happens when the guardrail service times out or errors: does the request proceed unchecked, or is it blocked? This is the worst-documented field in the entire catalogue — only two vendors state it plainly, which means most teams are running a control whose failure behaviour they cannot know.	Not published	Not published
Fit & integration
Work to try itThe shape of the work on the vendor’s own quickstart, from swapping one base URL through to deploying infrastructure. An ordinal class rather than a duration, because elapsed time depends on accounts and quota we cannot see.	Change one base URL	Install a package
Work to run itThe same scale applied to the vendor’s recommended production path. For several products this is much heavier than the quickstart, which is exactly why both are recorded.	Change one base URL	Install a package
Can you self-host it todayWhether an install command is actually published. Several products advertise self-hosting while publishing no command to start from, which a plain yes/no would hide.	No self-hosting	No self-hosting
Works with the OpenAI SDKWhether an existing OpenAI-compatible client can be pointed at it by changing the base URL and key.	Yes	Yes
Vercel AI SDK supportWhether a first-party AI SDK provider package exists, or only a community package, a documented workaround, or the generic OpenAI provider pointed at a custom base URL.	Official provider package	Official provider package
Python framework integrationsLangChain, LangGraph, LlamaIndex and similar orchestration frameworks with a documented integration.	LangChain,LlamaIndex	LangChain,LlamaIndex
Callable from Cloudflare WorkersWhether the docs show calling this product from your own Worker. Deliberately separated from the several products whose own gateway runs on Workers, which is a fact about their infrastructure and not about your edge compatibility.	Their gateway runs on Workers, not yours	Not published
MCP supportWhich kind of MCP support this is: a gateway that governs many MCP servers, a hosted MCP server you connect to, MCP tools accepted inside the completion API, or client tooling. These are different products behind one acronym.	Hosted MCP server	MCP client tooling
Needs your own provider keyWhether an upstream provider account and key must exist before your first call works. A prerequisite rather than a step, and it can differ between the hosted and self-hosted forms of the same product.	Not needed	Not needed
Gate before models workWhether an enablement click, a quota grant, a paid tier or an approval form stands between a valid key and a working model call.	No gate	Free tier sees fewer models

The short version

OpenRouter

Choose it when Teams that want the broadest possible model and provider catalog behind one OpenAI-compatible key with unified billing.

Look elsewhere when You need self-hosting, air-gapped deployment, or a gateway that charges nothing on credit purchases.

Largest published catalog: 500+ models, 80+ providers
No token markup; provider list prices pass through
Public models endpoint with per-model pricing

5.5% credit-purchase fee (min $0.80) makes small top-ups expensive - $5 buys $5 of credit for $5.80
BYOK documentation describes two different meters ($25k list-price cost vs 1M requests)
No native content guardrails or PII redaction

Vercel AI Gateway

Choose it when Teams already shipping on Vercel with the AI SDK who want zero token markup and no credit-purchase fee.

Look elsewhere when You need the widest possible provider pool per model, self-hosting, or you are not otherwise on Vercel.

Zero markup on tokens, including with BYOK, and no credit-purchase fee
Public unauthenticated models endpoint with pricing and per-provider endpoint data
Per-request ZDR and provider allowlists at no extra cost

Narrower long-tail catalog and fewer provider deployments per open-weight model than OpenRouter
Team-wide ZDR and team-wide provider allowlists are metered at $0.10 per 1,000 requests
BYOK requires purchased credits, can silently fall back to Vercel credentials, and is excluded from budgets

Estimate what these cost you →

A capability table cannot tell you which is cheaper for your traffic. Feed in your own volume and see the arithmetic.

OpenRouter vs LiteLLM vs Vercel AI Gateway at a glance

The three products people most often weigh against each other, on the twelve fields that usually decide it. Use the tool above for any other combination, or for all 113 fields.

Summary comparison of OpenRouter, LiteLLM, Vercel AI Gateway across pricing, deployment, and compliance fields
Attribute	OpenRouter	LiteLLM	Vercel AI Gateway
What kind of product	Managed marketplace	Open source	Managed gateway
Who runs it	Managed only	Self-host only	Managed only
Models available	458	Not published	386
Markup on model prices	None	None	None
Fee to add funds	5.5%	None	None
Monthly cost per person	Not published	None	None
Can use your own provider accounts	Yes	Yes	Yes
Does not retain your data	Yes	Not applicable	Yes
Will sign a HIPAA agreement	Not published	Not published	Yes
SOC 2 audited	Yes	Not published	Yes
Prompt cache offered	Passes provider caching through	Both exact and semantic	No gateway-owned cache
Licence	Proprietary	MIT	Proprietary

"Not published" means the vendor has not stated a figure. It is not a No. Model counts are not comparable across vendors, because counting conventions differ — treat them as an order of magnitude, not a score. Every value here is sourced on the provider pages, and the methodology explains how it was checked.

Questions about this table

Why are some cells in the comparison empty?

A blank means the vendor does not publish that fact, not that the answer is no. Every value in this table is read from a vendor's own page and stores the URL it came from, so there is nothing to show when no page states it. Treat a blank as a question to ask on a sales call rather than as a mark against the product.

Why can you only compare four products at a time?

Four columns is what stays readable on a laptop screen across 113 fields without horizontal scrolling doing the comparing for you. All 31 products are selectable and the set is swappable, so a wider shortlist is a sequence of four-way reads. The full list, grouped by category, is on the products page.

How current are the figures in this comparison?

Each value carries the date it was last checked, and a field is flagged once it passes the staleness window without re-verification. The changelog records every edit with its previous value and the source that justified it. Where a figure here differs from a vendor's page, the vendor's page is newer and the correction form is the fastest fix.

Pairs we have written up

The tool above compares any combination. These 46 pairs also come with a verdict, the single question that decides them, and the traps to watch for.

Velokey OpenRouter Does your workload need provider-key flexibility or consolidated chat and media access?
Velokey Eden AI Does your workload need a broader task API or consolidated chat and media access?
AI Gateway HQ Portkey Do you need a hosted governance boundary or the option to run the gateway yourself?
AI Gateway HQ LiteLLM Who should operate the gateway and own its deployment boundary?
AI Gateway HQ Respan Do you primarily need operating controls or an integrated tracing and evaluation workflow?
OpenRouter Vercel AI Gateway Do you need the long tail of the model catalog, or do you already pay Vercel for seats?
OpenRouter Requesty Do you bring your own provider keys, and do you need a completed SOC 2 attestation today?
LiteLLM Portkey Do you want to build and run the governance layer yourself, or pay for one that already exists?
LiteLLM Bifrost Do you value provider breadth and a large operator community, or a Go runtime without a documented throughput ceiling?
Portkey Helicone Are you starting something new, or do you already run Helicone and are deciding whether to stay?
Helicone Braintrust Gateway Do you want to measure production traffic after the fact, or gate quality before you ship?
Amazon Bedrock Google Vertex AI Which cloud already holds your data, and do you need FedRAMP High or explicit per-feature ZDR documentation?
Amazon Bedrock Azure AI Foundry Do you need first-party OpenAI models and catalog breadth, or AWS-native governance and FedRAMP High?
Cloudflare AI Gateway Vercel AI Gateway Which platform do you already deploy on, and do you need documented gateway-level compliance?
Groq Together AI Is perceived output speed your product differentiator, or do you need catalog breadth, fine-tuning and a HIPAA BAA?
Fireworks AI Together AI Do you need selectable latency tiers and reinforcement fine-tuning, or broader modality coverage and cheaper supervised fine-tuning?
Kong AI Gateway Apache APISIX AI Gateway Do you need LLM-native features and compliance paperwork, or is free and fast with basic proxying enough?
OpenRouter LiteLLM Do you want one commercial relationship and one bill, or do you want to hold every provider account and key yourself?
OpenRouter Portkey Do you need to reach the most models, or to control how your team uses the ones you already reach?
LiteLLM Cloudflare AI Gateway Do you want the gateway inside your infrastructure, or at an edge you never have to run?
Vercel AI Gateway LiteLLM Are the seats already on your invoice, and can you tolerate a gateway you cannot self-host?
Portkey Kong AI Gateway Is this a new AI product to adopt, or one more route on the API gateway your platform team already operates?
Amazon Bedrock OpenRouter Does the model call have to stay inside your AWS account and its compliance boundary?
LLM Gateway Requesty Do you want a fee you can escape by self-hosting, or a hosted router with EU residency that you must trust with full request logs?
LLM Gateway Bifrost Is a permissive licence with no hosted escape hatch worth more to you than a copyleft one that comes with a managed service?
TrueFoundry AI Gateway Orq.ai Router Do you need the complete compliance paperwork, or the widest reach and request-tier pricing?
TrueFoundry AI Gateway Apache APISIX AI Gateway Do you want a free foundation project you operate entirely yourself, or a paid platform with a vendor and an audit report behind it?
Orq.ai Router Braintrust Gateway Is the surrounding platform an eval workbench or a governance layer, and does the gateway underneath it reach far enough?
Groq Fireworks AI Do you need maximum output speed on a handful of models, or a broad open-weight menu with a business associate agreement?
Azure AI Foundry Google Vertex AI Which cloud are you already in — and if that is genuinely open, does catalog breadth or the tighter logging default matter more?
New API LiteLLM Are you charging the people who use the gateway, or standardising provider access for your own platform?
Eden AI OpenRouter Is a meaningful share of your workload non-LLM — OCR, speech, document parsing — or is it all chat?
Envoy AI Gateway Kong AI Gateway Do you need a vendor who can sign a contract and an SLA, or a project you own outright with nothing gated behind a quote?
Higress Apache APISIX AI Gateway Is the deciding factor the depth of the AI plugin shelf, or request-path reliability and framework integrations that work without you assembling them?
MLflow AI Gateway Braintrust Gateway Which evaluation platform are you already running, and can you accept its gateway on the terms it comes with?
Envoy AI Gateway Higress Do you need the wider plugin shelf, or upstream credentials your security review will accept and a control-plane API that will not move?
agentgateway Envoy AI Gateway Are you adding LLM routing to an Envoy Gateway you already run, or do you want one standalone data plane that also carries agent protocols and content guardrails?
agentgateway Bifrost Do you want the whole feature set in one Apache-2.0 build from a foundation project, or a generous free build from one vendor with guardrails, clustering and SSO behind a quote?
Eden AI Requesty Would you rather pay a fee on money in or a fee on tokens out, and which retention model can you actually sign?
New API LLM Gateway Do you want to run the gateway yourself and bill the people behind it, or hand the running of it to someone with an attestation?
MLflow AI Gateway Helicone Do you need per-user request analytics badly enough to adopt a platform whose owner is offering migration help?
Merge Gateway Vercel AI Gateway Is 5% of every token worth paying for content-level controls, or is a fixed seat fee with nothing on tokens the better shape?
Merge Gateway Portkey Are you buying a governance platform to run against your own provider keys, or buying model access with governance attached?
OpenRouter Hugging Face Inference Providers Is every model you need open-weight, and are you willing to bring your own guardrails, cache and compliance story to save the credit fee?
Eden AI Hugging Face Inference Providers Is your work beyond chat about reading documents and audio, or about generating images and video from open-weight models?
Together AI Hugging Face Inference Providers Do you need the things only the destination sells — fine-tuning, dedicated GPUs, batch, per-request controls — or do you need to not be tied to one destination?

Interactive comparison needs JavaScript

Each provider page carries the same full specification in a static table, and the public JSON API returns the whole dataset if you would rather compare it yourself.