OpenRouter is the best-known LLM router: one key, hundreds of models, provider list prices and automatic fallback. Our earlier post, OpenRouter alternative for agents: flat-rate vs per-token, looks at one question in depth: when a flat plan beats a router for an always-on agent. This page takes a wider view. It is a buyer’s table across six services, so you can match the alternative to the job.
What OpenRouter does well
From OpenRouter’s own pages, checked 2026-11-14.
- No inference markup. “Inference is billed at the provider’s list price on every plan.”
- Choice. Hundreds of models from many providers behind an OpenAI-compatible API, including “25+ free models”.
- Fallback and routing across providers serving the same model.
- Data controls. “Prompt and completion are not logged by default,” and zero data retention can be required per request with
"zdr": trueor set account-wide. - No platform cap on paid models. Paid variants have “no platform-level request cap”, though providers may still return 429s.
- EU in-region routing on Business and Enterprise plans.
Why people look for an OpenRouter alternative
From OpenRouter’s own pages.
- The credit fee. “OpenRouter’s fee is charged when you buy credits, 5.5% on Standard and 8% on Business.” The FAQ adds a USD 0.80 minimum per purchase. At high spend, that percentage is real money.
- BYOK fee at scale. Bring-your-own-key use is free up to “USD 25,000 of list price inference / month”, then “5% fee after”.
- Free-tier ceilings. Free variants allow 20 requests a minute, and 50 a day until you have bought 10 credits.
- Balance lockout. With a negative balance “you may see” errors, “including for free models”.
- Upstream rate limits. Paid limits are the provider’s; a busy provider means 429s you can’t buy your way out of on OpenRouter.
- Per-token only. No flat plan, so an agent’s bill grows with every retry.
- Privacy depends on settings. Many providers sit behind the router with different policies. ZDR exists, but you have to turn it on.
The buyer’s comparison table
| OpenRouter | Requesty | LiteLLM | Together AI | DeepInfra | BroadNet Inference | |
|---|---|---|---|---|---|---|
| Type | Hosted router | Hosted gateway | Self-hosted gateway | Direct host | Direct host | Direct host |
| Fee on top of model price | 5.5% on credits (8% Business) | 5% on top-ups | None (you run it) | None | None | None |
| Pricing model | Provider list price per token | Provider price per token | Each provider’s own | Per token; GPUs per hour | Per token or execution time | Per token, per audio minute, flat plans |
| Flat plan | No | No | Not applicable | No | No | Yes: unlimited DeepSeek V4.1 Flash, $60/month per session |
| OpenAI-compatible | Yes | Yes | Yes | Yes | Yes, with caveats | Yes |
| Data retention (cited) | Not logged by default; ZDR per request or account | Payload only if logging on; metadata always | Your servers, your settings | No prompt or response retention if you choose ZDR | Not stored on disk; deleted from memory | “keeps no prompts, messages, images, audio, responses or transcripts” |
| Model range | Hundreds | 600+ from 30+ providers | Any provider you add | Dozens serverless | Very wide | Six: four chat, two speech-to-text |
| Free credit | None stated; 25+ free models | Free plan, 200 requests/day | Free software | None; USD 15 first purchase | None; card or prepay | $5, no card |
| Cross-provider fallback | Yes | Yes | Yes | No | No | No |
Sources for every cell are listed at the end of the page.
What the fees mean in money
Fees are percentages, so they scale with spend. For a team that buys USD 1,000 of model usage a month:
| Service | Fee rule | Fee on USD 1,000 |
|---|---|---|
| OpenRouter Standard | 5.5% on credit purchases | USD 55 |
| OpenRouter Business | 8% on credit purchases | USD 80 |
| Requesty | 5% on top-ups | USD 50 |
| LiteLLM, Together AI, DeepInfra, BroadNet | No platform fee | USD 0 |
The fee is not the whole story. LiteLLM costs engineering time and a server. Direct hosts have no fee but also no fallback across providers. And the model price itself usually matters far more than a 5% fee: the same open model can be listed at very different prices by different hosts, as the provider pricing comparison shows.
9 alternatives to OpenRouter at a glance
Each line is from the provider’s own pages, checked 2026-11-14.
- Requesty: hosted gateway for 600+ models from 30+ providers, with a 5% fee on top-ups.
- Vercel AI Gateway: hosted gateway that charges “no markup and no platform fee on tokens”, including with your own keys. Payment processing fees and optional add-ons, such as 5% for regional endpoints, are extra.
- Cloudflare AI Gateway: analytics, caching and rate limiting are free on all plans; its unified billing adds 5% to credit purchases.
- Eden AI: about 1,000 models from 80+ providers, with a 5.5% fee when you buy credits and no subscription.
- Portkey: gateway with a free Developer tier (10,000 logged requests a month), Production at USD 49 a month, and a self-hosted open-source version.
- LiteLLM: open-source proxy you run yourself, covering 100+ providers with no platform fee.
- Together AI: direct host for open models, fine-tuning and GPUs; the first purchase is at least USD 15.
- DeepInfra: direct host with the lowest discounted DeepSeek V4.1 Flash prices we checked and 200 concurrent requests per model.
- BroadNet Inference: direct host of six models with flat unlimited plans ($60 a month per session) and zero retention. This is BroadNet’s own product.
Pick by job
| Your job | Best fit | Why |
|---|---|---|
| Try many models this week | OpenRouter | Hundreds of models, free variants, one key |
| Gateway with team budgets, hosted | Requesty | Per-key budgets and analytics, 5% fee |
| Gateway with no platform fee | LiteLLM | MIT-licensed; you run it (see the LiteLLM proxy guide) |
| Open models, fine-tuning, GPUs | Together AI | Serverless plus dedicated GPUs and training |
| Low per-token prices on open models | DeepInfra | Lowest discounted prices we checked on DeepSeek V4.1 Flash |
| One agent running all day on one model | BroadNet Inference | Flat plan, no token meter |
| No retention without configuration | BroadNet Inference | Nothing retained on any request |
| Arabic speech to text | BroadNet Inference | Dedicated Arabic model, rank reported by BroadNet |
When to stay with OpenRouter
- You change models often or need models no single host serves.
- You want automatic fallback when a provider is slow or down.
- Your spend is moderate, so 5.5% is small next to the value of one key.
- You need EU routing and are on Business or Enterprise.
When BroadNet Inference fits
BroadNet Inference is a single provider, not a router, and has no cross-provider fallback. It serves six models under zero data retention. Its privacy statement says its own and hosted open-weight models run “on BroadNet-owned or leased hardware in the United States”, and that partner-served models, such as DeepSeek V4.1 Flash, run “under zero-data-retention terms with our US inference partners”. It fits when:
- One model does most of the work all day. Unlimited DeepSeek V4.1 Flash is $60 a month, $15 a week or $1 an hour per session. The flat-rate vs per-token calculator shows the break-even for your usage.
- You want zero retention by default. See zero data retention.
- You need a 1M-token context or Arabic transcription on the same key.
Per token, BroadNet does not win on price: DeepInfra’s current discounted DeepSeek V4.1 Flash prices are lower. Its case is the flat plan and the retention policy.
When another alternative fits better
- Requesty if you like the hosted-gateway model and want per-key budgets; its fee is 5% on top-ups.
- LiteLLM if you have the people to run a proxy. It removes the platform fee and keeps logs on your servers.
- DeepInfra if price per token decides, and Together AI if you need fine-tuning or dedicated GPUs.
How to switch from OpenRouter
Change the base URL, key and model ID. OpenRouter IDs are provider/model; BroadNet IDs start with broadnet/.
from openai import OpenAI
import os
client = OpenAI(
base_url="https://inference.broadnet.ai/v1", # was https://openrouter.ai/api/v1
api_key=os.environ["BROADNET_API_KEY"],
)
resp = client.chat.completions.create(
model="broadnet/deepseek-v4.1-flash", # was deepseek/deepseek-v4.1-flash
messages=[{"role": "user", "content": "List the risks in this migration plan."}],
max_tokens=2000,
)
- Pull a month of usage from OpenRouter’s activity page, grouped by model.
- For the top model, price it with the LLM API cost calculator, adding the 5.5% credit fee to the OpenRouter side.
- Check concurrency: BroadNet pay-as-you-go keys allow 30 requests a minute and 4 in flight; the rate limit calculator helps size it.
- Keep OpenRouter for everything else. The provider checklist covers the remaining questions.
The short version
OpenRouter is the easiest way to reach many models. Its alternatives each fix one thing: Requesty and LiteLLM change who runs the gateway and what it costs, Together AI and DeepInfra remove the middle layer, and BroadNet Inference replaces the meter with a flat plan. Many teams use two of these at once.