AI Router
Login

Unlimited AI Access. One Flat Price.

smart_toy OpenClaw Friendly
public Swiss-Hosted
Qwen3.8
DeepSeek DeepSeek-V4
auto_awesome OpenAI Compatible

Stop counting tokens. Just build.

Access unlimited Qwen3.8, DeepSeek-V4 & more for a flat CHF 39/mo.
Full privacy — no prompt logging, no training on your data, just reliable low-latency AI access.

Perfect for vibe coding sessions and 24/7 agents — no token counting (fair-use).

🚀 NOW LIVE — Instant access available

Start using AI Router in seconds.

Instant API key • Cancel anytime

savings Flat rate — no token counting
schedule 24/7 agents — continuous workloads
speed Low latency — Swiss-hosted
shield Privacy — no prompt logging

Flat Rate Pricing

CHF 39/mo

No hidden fees. No token counting.

Context length

262K

Tokens context window — massive capacity.

Compatibility

100%

Drop-in replacement for OpenAI.

Developer Friendly.
Built for production workloads.

Integration takes minutes, not days. We maintain full compatibility with the OpenAI SDK, so you can switch your base URL and API key to start saving immediately.

terminal

OpenAI Compatible

Drop-in replacement for your existing client. Just change the base URL.

bolt

High Throughput

Dedicated capacity ensures consistent latency and tokens per second.

library_books

262K Context

262K token context window for RAG and document processing.

Drop-in replacement. Same SDK. Same calls. No migration.

{
"models": {
"providers": {
"airouter": {
"baseUrl": "https://api.airouter.ch/v1",
"apiKey": "${AIROUTER_API_KEY}",
"api": "openai-completions",
"models": [
{
"id": "Qwen3.8",
"name": "Qwen3.8 (airouter.ch)",
"contextWindow": 262144,
"maxTokens": 65536,
"reasoning": true,
"input": ["text", "image"],
"cost": { "input": 0, "output": 0 },
"compat": {
"supportsUsageInStreaming": true,
"supportedReasoningEfforts": ["none", "low", "medium", "xhigh"]
}
}
},
{
"id": "DeepSeek-V4-Flash",
"name": "DeepSeek-V4-Flash (airouter.ch)",
"contextWindow": 262144,
"maxTokens": 65536,
"reasoning": true,
"input": ["text", "image"],
"cost": { "input": 0, "output": 0 },
"compat": {
"supportsUsageInStreaming": true
}
}
}
]
}
}
},
"agents": {
"defaults": {
"model": {
"primary": "airouter/Qwen3.8",
"fallbacks": ["airouter/DeepSeek-V4-Flash"]
}
}
}
}

Why AI Router Switzerland

AI Router Switzerland is designed for developers and AI enthusiasts who want to focus on building, testing, and running AI workflows without worrying about token limits or overages. Our "unlimited" API means you can:

  • Run long coding sessions or 24/7 agents without interruptions.
  • Integrate AI into local tools, IDEs, or autonomous agents seamlessly.
  • Enjoy Swiss-hosted privacy — no prompt logging, no training on your data. Only light metadata analysis is performed to ensure consistent performance for everyone.

Combined with generous operational limits (3 parallel requests, 240 requests/min, 10M tokens/min), low-latency infrastructure, and OpenAI-compatible APIs, AI Router provides a reliable and worry-free environment for experimentation, development, and production-grade agent workflows.

Unlimited Access Full Privacy Developer-Friendly Low latency

Available Models

Powerful AI models ready for production workloads.

Qwen3.8
Context length 262K tokens
Parameters 27B (dense)
Quantization Weights FP8
Quantization KV Cache FP8
Architecture Hybrid (DeltaNet + Attention)
Reasoning ✓
Tool calling ✓
Image input ✓
Created by Alibaba
Release August 14, 2026
HF Slug Qwen/Qwen3.8-27B-FP8
Code RAG Agents Reasoning Tool calling Vision 119 Languages

Best for

Agentic coding, repository-level reasoning, RAG, document analysis

Strengths

Agentic orchestration, repo-level coding, long-context workflows, production-ready stability

SWE-bench Pro

Real-world software engineering

61.7%

GPQA Diamond

Graduate-level scientific reasoning

89.2%

QwenSWEBench

Software engineering (avg@3, 8h timeout)

79.0%

Humanity’s Last Exam

Multi-disciplinary research evaluation

30.8%

LiveCodeBench v6

Real-world coding benchmark

90.3%

Agents' Last Exam

Frontier agentic tasks (Pass@1)

20.4%

Terminal Bench 2.1

Agentic terminal coding

73.0%

IFBench

Instruction following

79.5%

The gold standard for open-weight models. Qwen3.8-27B brings a unique hybrid architecture combining Gated DeltaNet memory with traditional attention, giving it superior agentic coding and repository-level reasoning. With thinking preservation across conversation turns and support for 119 languages, it's built for developers who need stability and real-world utility.

DeepSeek
DeepSeek-V4-Flash
Context length 262K tokens
Parameters 284B (13B active)
Quantization Weights FP4 + FP8 Mixed
Quantization KV Cache FP8
Architecture MoE (CSA + HCA)
Reasoning ✓
Tool calling ✓
Image input ✓ augmented
Created by DeepSeek
Release July 31, 2026
HF Slug deepseek-ai/DeepSeek-V4-Flash-0731
Code Agents Reasoning Tool calling Vision 100+ Languages

Best for

Reasoning, coding, agentic tasks

Strengths

Fast MoE inference, top coding & reasoning benchmarks, cost-efficient deep reasoning

DeepSWE

Original long-horizon software engineering

54.4%

GPQA Diamond

Graduate-level scientific reasoning

88.1%

SWE-bench Verified

Real-world software engineering

79.0%

Humanity’s Last Exam

Multi-disciplinary research evaluation

34.8%

LiveCodeBench v6

Real-world coding benchmark

91.6%

Agents' Last Exam

Frontier agentic tasks (Pass@1)

25.2%

Terminal Bench 2.1

Agentic terminal coding

82.7%

HMMT 2026 Feb

Mathematical problem solving

94.8%

DeepSeek-V4-Flash is a 284B Mixture-of-Experts model that activates just 13B parameters per token, delivering frontier reasoning and coding with highly efficient inference. Its hybrid CSA + HCA attention architecture and MoE design make it exceptionally fast on agentic workloads, while deep thinking mode provides thorough reasoning for complex problems. If you need raw benchmark performance, this is the pick.

Embedding, Speech-to-Text & Text-to-Speech

Qwen3-Embedding
Context length 32K tokens
Parameters 4B (dense)
Quantization Weights Q6_K
Quantization KV Cache Q8_0
Architecture Decoder-Only Transformer
Embedding Dimension 2560
Supported Languages 100+
Created by Alibaba
Release 2025
HF Slug Qwen/Qwen3-Embedding-4B-GGUF
Embeddings RAG Semantic Search

Best for

Agent memory indexing, RAG pipelines

Strengths

Semantic search, code retrieval, knowledge base indexing

State-of-the-art text embedding model designed for retrieval, ranking, and similarity tasks. With 2560-dimensional vectors, 32K context length, and support for 100+ languages including programming languages, it excels at text retrieval, code retrieval, classification, and clustering.

whisper-large-v3-turbo
Context length 25 MB
Parameters 809M
Quantization int8_float16
Speed ~50× realtime
Architecture Encoder-Decoder
Input mp3, wav, m4a, ogg, webm
Supported Languages 99+
Created by OpenAI
Release 2024
HF Slug openai/whisper-large-v3-turbo
Speech-to-Text Multilingual Transcription

Best for

Real-time transcription, voice agents, meeting notes

Strengths

Robust to noise & accents, 99+ languages, zero-shot transcription

OpenAI Whisper large-v3-turbo running on dedicated GPU infrastructure. Low-latency speech-to-text with broad language support and high accuracy across domains.

volume_up
kokoro
Context length 1K tokens (~5 min audio)
Parameters 82M
Quantization FP32
Speed ~74× realtime
Architecture StyleTTS 2 + ISTFTNet
Output mp3, wav, opus, flac, aac, pcm
Supported Languages 10 (68 voices)
Created by Hexgrad
Release January 2025
HF Slug hexgrad/Kokoro-82M
Text-to-Speech 68 Voices Audio replies

Best for

Voice assistants, audio replies, narration, accessibility

Strengths

~74× realtime, 68 natural voices, arbitrary-length input

Kokoro text-to-speech. 68 natural voices across 10 languages.

What's Included

✓ Unlimited API requests
✓ Swiss-hosted infrastructure
✓ Qwen3.8 + DeepSeek-V4
✓ OpenAI-compatible API
✓ Low latency
✓ Embeddings + Whisper STT + Kokoro TTS

Frequently Asked Questions

Ready to unleash unlimited intelligence?

Subscribe today and start building.