Top LLM gateways and routers compared

About Top LLM gateways and routers compared

An LLM gateway sits between your code and a scatter of model providers, and the choices that matter come down to billing shape, provider reach and how you get help when a request fails. This ranking covers 42 gateways and routers, from open-source proxies you host yourself to paid platforms that resell model access. The central trade-off is between bringing your own key, which keeps provider bills direct and gateway charges near nothing, and using a platform's own credits, which simplifies the invoice but adds a markup or platform fee to every token. Entry price and renewal diverge as well: several gateways here open with a free tier or a free self-hosted option and then settle into metered token charges or a per-model monthly charge once traffic grows, while a few bill from the outset with no free tier at all. Provider reach varies widely, from routers fronting from 4 providers to LiteLLM at 140. Most of these options publish an OpenAI-compatible API, so moving between them rarely means rewriting client code, and the typical gateway in this ranking routes 9 providers.

How we picked

Every score here is built from three weighted inputs: cost at 40%, offer at 40% and trust at 20%. Cost and offer together carry most of the weight, so the money you pay at entry, the money you pay at renewal and what a plan actually includes decide most of the order, with trust breaking the remaining ties. Cost is read from the fee model each vendor publishes: whether self-hosting stays free, whether the gateway adds a markup or a flat platform fee to provider list rates, whether pay-as-you-go credits need a minimum top-up, and whether a free tier exists. Offer is read from plan contents: providers routed, whether an OpenAI-compatible API is published, and which support channels are offered. Trust is read from the contracting legal entity, headquarters country, founding year and the support routes a vendor lists. Figures come from vendor pricing pages, terms and conditions, provider and model listings, and corporate or encyclopaedia records for founding dates. Partners that serve your country are listed first in this ranking, then the rest by score. Close scores break toward the wider provider reach.

Our top picks

1. 3. Nexos.ai: best for Nexos.ai

   nexos.ai offers an AI gateway giving one unified API access point to 200+ models, with routing, caching, failover and budget controls.

   • Headquarters: Lithuania
   • Founded: 2024

2. 4. Salad

   Strong on billing shape rather than breadth, which is why it sits near the top of this ranking. Salad offers a flat monthly subscription or pay per token, the only flat monthly figure among these options, and it publishes an OpenAI-compatible API. The trade-off is control: it does not take your own key and it has no free tier, so the account starts paid and every renewal follows the same plan. Support runs on email, help centre and community, and the contracting entity is Salad Technologies, Inc., founded 2018. UCloud, the member below it, offers an OpenAI-compatible API across from 4 providers from Hong Kong rather than a fixed monthly price.

   • GPU models: NVIDIA RTX 3060, NVIDIA RTX 3070, NVIDIA RTX 3070 Ti, and 6 more
   • GPU hour from: $0.015
   • Billing granularity: Per second

   Pros

   • Billing is per second against an hourly model, so short runs and interrupted jobs pay only for the time they hold a card.
   • Workloads sit in from 190 locations, well above the 6 median in the GPU ranking.
   • The catalogue in NVIDIA RTX 3060, NVIDIA RTX 3070 and NVIDIA RTX 3070 Ti spans several RTX generations, the RTX PRO workstation class and selected AMD classes.
   • The AI gateway is OpenAI-compatible, so clients already written against that API can be repointed at Salad.
   • The operator is Salad Technologies, Inc., a US company founded in 2018, with SOC 2 Type I listed for security.

   Cons

   • One GPU per node: None - single-GPU nodes, no multi-GPU NVLink interconnect, so any plan that needs several accelerators inside a single machine has to look elsewhere.
   • Bring your own key is not offered, and the gateway is billed as Flat monthly subscription or pay per token, so you cannot pay through your own provider account.
   • Support runs on email, help centre and community only, with no live chat or phone line listed for urgent issues.
   • A free tier is not offered, and only 27% of the GPU ranking sits in the same paid-only position.

3. 5. AI/ML API

   Model choice is the reason it holds a place in this ranking. The provider count here is from 1,000, the widest in the group by a wide margin, reached through pay-as-you-go per-model pricing on a prepaid balance. AI/ML API publishes an OpenAI-compatible API and supports on email, chat and help centre, including live chat. It has no free tier, so an account starts funded, and the contracting entity is Boiler Labs FZ-LLC in the United Arab Emirates, which matters for teams that filter vendors by region. Requesty, the member above, keeps a free tier and a far smaller catalogue, while SiliconFlow below does the opposite.

   • Input price from: $0.0516 to $41.26
   • Output price from: $0.138 to $206.31
   • Models offered: GPT-5.6 Sol, GPT-5.5, GPT-5, and 34 more

   Pros

   • One OpenAI-compatible endpoint fronts from 1,000 providers, so a client can move between vendors by changing the model name.
   • The catalogue in GPT 5.6 Sol, GPT 5.5 and GPT 5 mixes frontier text models with image and video endpoints such as FLUX 3 Video and Kling 3.0 Pro, broader than the single-vendor set at gpt 6 astra, gpt 6 luna and gpt 6 sol.
   • A context window of 1,050,000 tokens ranks 5 of 72 tracked brands, so long documents travel in one request.
   • The entry input range, $0.052 to $41.26, opens below Salad's $0.06, so a token-light trial is cheaper here while the top of the range runs well above it.

   Cons

   • Free tier: not available, so the catalogue cannot be sampled at no cost and a prepaid balance has to be funded before the first call.
   • Billing is pay-as-you-go per model against a prepaid balance, and the fee model sets a minimum top-up, so spend follows usage rather than a tier you pick.
   • On price the brand reads mid-field: entry input ranks 32 of 71 and entry output 26 of 70.

4. 6. SiliconFlow

   The media-plus-text option in this ranking. SiliconFlow bills pay-as-you-go by token or by image and video, postpaid and with no minimum commitment, offers a free tier, and routes 6 providers through an OpenAI-compatible API. Support runs on email, help centre and community, and the contracting entity is SILICONFLOW LABS PTE. LTD. in Singapore, so teams with regional requirements have a clear contracting party. Fireworks AI, the member below it, is broader in deployment style, charging per token for serverless work and per GPU second or hour for on-demand and dedicated deployments, where SiliconFlow sticks to token and media pricing.

   • Input price from: $0.040
   • Output price from: $0.18
   • Models offered: DeepSeek-V4.1-Flash, DeepSeek-V4-Flash-Vision-Exp, DeepSeek-V4-Pro-0813, and 26 more

   Pros

   • A free tier with $1 in free credits in credit runs real prompts on the live catalog before spend begins.
   • Entry pricing of $0.04 on input and $0.18 on output sits below Salad, whose input rate starts at $0.06.
   • 6 providers route through the same account, so a fallback model is a configuration change rather than another supplier to onboard.
   • The OpenAI-compatible API lets clients written for that interface point at SiliconFlow by changing the base URL rather than rewriting requests.
   • NVIDIA H100, NVIDIA H200 and AMD MI300 cover NVIDIA and AMD parts, and BYOC (Bring Your Own Cloud) deployment suits teams that keep compute in a cloud they already run.

   Cons

   • SiliconFlow's input price of $0.04 ranks 21 of 71 brands, so the entry rate is competitive rather than the lowest in the field.
   • Output pricing of $0.18 ranks 30 of 70, and output cost is where the bill grows fastest on long answers.
   • The catalog leans on open-weight families rather than the closed names Azure AI Foundry lists at gpt 6 astra, gpt 6 luna and gpt 6 sol, which narrows the shortlist for a team set on those models.
   • Support runs on email, help centre and community only, which suits self-serve debugging over a phone conversation with an account manager.

5. 8. UCloud

   In this ranking for a broad-enough route table and a long operating history. UCloud offers an OpenAI-compatible API across from 4 providers, is headquartered in Hong Kong, and was founded 2012, earlier than either member above it. Support runs on email and help centre, email and a help centre, and the contracting entity is UCLOUD INFORMATION TECHNOLOGY(HK) LIMITED. For a team that needs routing, failover and a single endpoint across a small provider set, that is enough. Supascale AI, the member below it, adds a free tier and a per-user private network, where UCloud's draw is the longer track record and the larger catalogue.

   • GPU models: RTX 50 Series, RTX 40 Series (48G), RTX 40 Series, and 7 more
   • Billing granularity: Hourly
   • Free credits: $5 credit on contacting support after linking a credit card

   Pros

   • The Kubernetes control plane is free, so managed clusters cost nothing beyond the nodes.
   • Certifications listed include CMMI Level 5, ISO 9001 and ISO 27001, which shortens vendor security review.
   • 28 locations across Asia, Europe, the Middle East and the Americas.
   • Block storage prices $0.063 to $0.24 per month, and the object tier follows the same bucket interface most tools already use.

   Cons

   • Support runs on email and help centre, so urgent issues go by email and the help centre.
   • The card list, RTX 50 Series, RTX 40 Series (48G) and RTX 40 Series, stops short of the newer accelerators that SberCloud and ComputePrices.com list.
   • Only the GPU range bills hourly; VPS and bare metal run on monthly terms.

What to look for

Cost at entry and at renewal

Typical input pricing in this category is €0.04 and typical output pricing €0.16, with the widest input price at €0.83. Read the rate that opens an account next to the rate that renews each month, because the two are rarely the same figure.

Bring your own key

70% of listings here accept your own provider keys, against 30% that do not. This is the clearest split on price control, since your own key keeps the provider relationship and its invoice in your hands.

How many providers it fronts

The typical gateway in this category routes 9 providers, the upper quartile reaches 19, and the widest sits at 140.

Billing model and granularity

Hourly billing is the norm at 80%, per-second billing at 20%, and hourly models make up 71.4% of listings against 42.9% on a free tier.

Free entry versus free renewal

A free tier appears on 78.8% of listings, yet a free entry price rarely means free ongoing use, so check what the plan caps and what the renewal rate is.

Support you can actually reach

Email, a help centre and a community are each common at 85.4%, 82.9% and 85.4%, while live chat is rarer at 31.7% and phone support rarest at 19.5%.

Frequently asked questions

Top LLM gateways and routers compared by country

• Brazil
• Canada
• France
• Germany
• India
• Italy
• Japan
• Mexico
• South Korea
• Spain
• United Kingdom
• United States
• Vietnam

Providers compared head to head

• BytePlus vs WaveSpeedAI
• BytePlus vs Nexos.ai
• BytePlus vs Salad
• WaveSpeedAI vs Nexos.ai
• WaveSpeedAI vs Salad
• Nexos.ai vs Salad