OpenRouter alternatives (2026)—cheaper, multimodal, your own keys, and the EU question

People look for an OpenRouter alternative for one of four reasons: to pay less per token (OpenRouter passes through provider list prices and charges a 5.5 % fee, with a 0.80 $ minimum, on every credit purchase by card), to call image, video, and music models with the same key, to keep your own provider keys behind a control layer you operate — or to have requests processed in the EU, the question that comes up most often in French search results. This comparison matches each reason with the tool that addresses it, including cases where staying with OpenRouter is the right choice.

The short answer: to pay less for Claude, Kunavo lists all Claude models 30 % below Anthropic's list price (Claude Sonnet 5 at $1.40 / $7.00 per million input/output tokens, versus $2.00 / $10.00). For self-hosted open source, LiteLLM; for routing and observability on your own keys, Portkey or Helicone; for open-weight models at scale, Together AI. If EU processing is a contractual requirement, Kunavo is not the answer — details below.

Disclosure up front: Kunavo is our product, so it is listed first. Every claim about it is shown with figures on the Kunavo vs. OpenRouter page, and each other tool gets a genuine recommendation.

The short list

One distinction does most of the work: inference gateways (OpenRouter, Kunavo, AIMLAPI) resell model access through a single balance, while BYO-key tools (LiteLLM, Portkey, Helicone, Cloudflare) add a control layer over provider accounts you already have. The two can be combined, and many teams use one of each. The LLM gateway guide (English) explains the pattern.

What the same workload costs

Cost of 10 million input tokens and 1 million output tokens with Claude, without prompt caching. The OpenRouter column is the pass-through list price plus the 5.5 % fee on credit purchases by card.

With cache or batch processing, both columns change. Kunavo’s rates are recalculated from the catalog at each build; OpenRouter’s are the list prices from its model list, retrieved on September 30, 2026.

1. Kunavo — Claude below list price, multimodal on one key

Kunavo overlaps with OpenRouter on the chat side—Claude and GPT behind an OpenAI-compatible endpoint—and differs in three respects. Price: Claude is 30% below Anthropic’s list price, and Claude Sonnet 5 costs $1.40 / $7.00 (list price $2.00 / $10.00), with no top-up fees. Modalities: image (Nano Banana, GPT-Image-2), video (Veo 3), and music (Suno, absent from OpenRouter’s model list) are full endpoints using the same key. Payment: Stripe Checkout with cards, Apple Pay, Google Pay, Link, and local methods depending on the country; prepaid starting at $10, no subscription, and the balance does not expire.

The tradeoff is catalog breadth: Kunavo offers frontier models it can provide below list price, rather than hundreds of niche models. No free models, shared upstream capacity without a contractual SLA, and a single region in the eastern United States. Run your own numbers in the LLM cost calculator, or see the point-by-point comparison at Kunavo vs. OpenRouter.

2. LiteLLM — self-hosted open source

LiteLLM is a Python proxy that you run yourself: an OpenAI-compatible endpoint in front of your own provider keys, with model routing, per-key budgets, and logging. It is the default answer to “open-source alternative to OpenRouter,” with one important caveat: it resells nothing. You sign contracts with providers, pay them directly, and operate the proxy. Because you configure the endpoints yourself, you also decide where requests are processed. Details: Kunavo vs. LiteLLM.

3. Portkey — the hosted control layer for your own keys

Portkey is a hosted gateway over your own provider keys: conditional routing, failovers, retries, guardrails, and request analytics, with no infrastructure to operate. Provider billing remains direct; Portkey charges for the platform. Suited to teams that want to bring many internal applications under one policy layer. Details: Kunavo vs. Portkey.

4. Helicone — observability first

Helicone approaches the space through logging: connected as a proxy or through an asynchronous SDK, then per-request traces, cost attribution, caching, and evals across your existing provider traffic. If the problem is “we don't know where our LLM budget is going,” start here. Details: Kunavo vs. Helicone.

5. Together AI — open weights at scale

Together hosts open-weight models — Llama family chat, embeddings, image — on its own serverless and dedicated infrastructure. It is less a router than a place to run open models fast, and the right choice if your strategy relies on open weights rather than a frontier API. Details: Kunavo vs. Together AI.

6. AIMLAPI — the media-focused aggregator

AIMLAPI aggregates a broad catalog of text and media models behind one API and overlaps with OpenRouter in breadth and image/video coverage. Compare its pricing model and payment terms with your usage profile. Details: Kunavo vs. AIMLAPI.

7. Cloudflare AI Gateway — the edge proxy

Cloudflare's AI Gateway sits at the edge in front of provider keys you already have: response caching, rate limiting, retries, and analytics. It does not resell inference; it is a pure control layer, combinable with any inference source, including OpenRouter and Kunavo. Details: Kunavo vs. Cloudflare AI Gateway.

From France: data residency and payment

The question that dominates French search results is about a European alternative. The honest answer: Kunavo is not one. Kunavo's gateway runs in a single region in the eastern United States and forwards requests from there to model providers. If EU processing is a contractual requirement, a BYO-key gateway operated in your own infrastructure is a better fit — for example, LiteLLM in front of the European endpoints specified in your provider contracts — because you decide where requests go.

Payment: OpenRouter’s FAQ lists major credit cards, AliPay, and cryptocurrency in USDC (as of September 30, 2026; PayPal is “coming soon”). Buying credits by card costs 5.5 % (minimum 0.80 $); USDC costs 5 %. Kunavo’s Stripe checkout accepts cards (Visa, Mastercard, Amex), Apple Pay, Google Pay, and Link, with no top-up fee; Bancontact is available in Belgium. Both set their prices in US dollars; Kunavo’s checkout displays the amount converted to local currency through Stripe, with a 2 to 4% conversion fee included in the rate and paid by the buyer. This fee can be avoided by paying in dollars.

Rates for all Claude models are in the Claude API pricing guide.

When OpenRouter remains the right choice

• You need the broadest catalog. Hundreds of models from more than 80 providers (OpenRouter’s pricing page, as of September 30, 2026): none of the alternatives listed here fully reproduces this breadth.
• You want free models. The :free variants and the openrouter/free router are limited to 50 requests per day, or 1,000 after purchasing $10 in credits (according to the FAQ). Kunavo has none.
• Your workload can wait. OpenRouter's model list offers :batch variants of Claude models at half the list price (checked September 30, 2026). For non-urgent batch processing, this is cheaper than any discount on real-time calls; Kunavo has no batch endpoint.
• You bring your own provider keys. On the pay-as-you-go plan, the first 25,000 $ of inference at list price per month is fee-free, then 5 % (same FAQ).
• You pay in USDC. OpenRouter accepts USDC with a 5 % fee; at Kunavo, crypto (stablecoins) appears only in the US Checkout in US dollars.

Switching takes two lines

Every serious alternative speaks the OpenAI protocol, so migration comes down to changing base_url and the key. Only the model ID changes: OpenRouter's provider-prefixed ID (anthropic/claude-sonnet-5) becomes the unprefixed ID (claude-sonnet-5).

import os
from openai import OpenAI

# OpenRouter, Kunavo et LiteLLM parlent le même protocole OpenAI
client = OpenAI(
    api_key=os.environ["KUNAVO_API_KEY"],   # avant : OPENROUTER_API_KEY
    base_url="https://api.kunavo.com/v1",    # avant : https://openrouter.ai/api/v1
)

resp = client.chat.completions.create(
    model="claude-sonnet-5",   # avant : anthropic/claude-sonnet-5
    messages=[{"role": "user", "content": "Hello"}],
)
print(resp.choices[0].message.content)

What a gateway should handle — routing, failover, caching, spending limits — is covered in the OpenAI-compatible API guide (English).

Frequently asked questions