Tensorix
Overview
Tensorix is an EU-sovereign AI inference platform providing private access to open-source language models. Founded in 2024 and registered in Ireland (Tensorix Ltd.), the company positions itself as a privacy-first, GDPR-compliant alternative to US-based AI providers. All infrastructure runs exclusively in Dublin and Helsinki data centres, avoiding US Cloud Act jurisdiction. They emphasize zero data retention, transparent compliance, and cost savings of 60-88% vs OpenAI.
Markets
- Enterprise AI inference — companies needing production LLM access with strong privacy guarantees
- EU-regulated industries — finance, healthcare, government requiring GDPR compliance and data residency
- Cost-sensitive teams — developers looking to reduce inference costs by switching from proprietary to open-source models
- Geographic focus: EU-first, particularly organisations subject to European data sovereignty requirements
Products
- Shared inference API — pay-as-you-go access to 14 confirmed LLMs + 2 audio models via OpenAI-compatible and Anthropic-compatible endpoints
- Audio API — TTS (chatterbox-turbo) and STT (faster-whisper-large-v3) via OpenAI-compatible endpoints
- Dedicated inference — isolated compute for enterprise customers (details limited on public site)
- Key differentiators: zero data retention, EU-only infrastructure, OpenAI SDK drop-in compatibility (change base_url only), publicly available DPA and sub-processor list
- Integrations ecosystem: 30+ documented integrations including Claude Code, Cursor, Cline, LangChain, LlamaIndex, Vercel AI SDK, n8n, Zapier, Dify, and more
Supported Models
| Provider | Model | Model ID | Context | Notes |
|---|---|---|---|---|
| Z-AI | GLM-5.1 | z-ai/glm-5.1 |
203K | Featured model, recommended for Claude Code |
| Z-AI | GLM-4.6 | z-ai/glm-4.6 |
203K | General purpose, bilingual |
| MiniMax | MiniMax-M2.5 | minimax/minimax-m2.5 |
197K | Reasoning, functions |
| MiniMax | MiniMax-M2 | minimax/minimax-m2 |
197K | Coding, fast |
| Moonshot AI | Kimi-K2.5 | moonshotai/kimi-k2.5 |
262K | Vision, functions |
| DeepSeek | DeepSeek-V3.1 | deepseek/deepseek-chat-v3.1 |
164K | General chat/coding |
| DeepSeek | DeepSeek-V3.2 | deepseek/deepseek-v3.2 |
164K | Fast responses |
| DeepSeek | DeepSeek-R1-0528 | deepseek/deepseek-r1-0528 |
164K | Complex reasoning |
| Alibaba | Qwen3-235B | qwen/qwen3-235b-a22b-2507 |
131K | Large MoE model |
| Alibaba | Qwen3-Coder-30B | qwen/qwen3-coder-30b-a3b-instruct |
262K | Coding specialist |
| Meta | Llama-3.3-70B | meta-llama/llama-3.3-70b-instruct |
131K | Instruction-tuned |
| Meta | Llama-4-Maverick | meta-llama/llama-4-maverick |
1050K | Largest context (1M+) |
| OpenAI | GPT-OSS-120B | openai/gpt-oss-120b |
131K | Open-source, reasoning |
| OpenAI | GPT-OSS-20B | openai/gpt-oss-20b |
131K | Open-source, lightweight |
| — | chatterbox-turbo | chatterbox-turbo |
— | TTS (text-to-speech) |
| Systran | faster-whisper-large-v3 | Systran/faster-whisper-large-v3 |
— | STT (speech-to-text) |
Last verified: 2026-04-12
Key Capabilities
| Capability | Status | Notes |
|---|---|---|
| OpenAI SDK compatibility | ✅ | Drop-in base_url swap |
| Anthropic SDK compatibility | ✅ | Separate endpoint: api.tensorix.ai |
| Streaming | ✅ | Standard SSE |
| Zero data retention | ✅ | Core differentiator |
| EU data residency | ✅ | Dublin + Helsinki only |
| GDPR compliance | ✅ | Irish-registered, public DPA |
| Dedicated inference | ✅ | Enterprise offering |
| Function calling | ✅ | Documented in API reference |
| Audio API (TTS/STT) | ✅ | chatterbox-turbo (TTS), faster-whisper-large-v3 (STT) |
| Caching | ❓ | Not documented |
| Fallback/routing | ✅ | "Intelligent routing and automatic failover" claimed |
| Rate limits | ✅ | 60 RPM, 2M TPM per key; enterprise custom limits |
| Usage dashboard | ✅ | app.tensorix.ai/dashboard/usage |
| Team accounts | ✅ | Documented in docs |
| Observability/logging | ❓ | Not documented beyond usage dashboard |
Last verified: 2026-04-04
Pricing
| Tier | Price | Notes |
|---|---|---|
| Pay-as-you-go | From $0.15/M tokens | No monthly fees, no minimums |
| TTS (chatterbox-turbo) | $0.000015/character | ~$0.15 per 10K chars |
| Dedicated inference | Custom | Enterprise, contact sales |
Last verified: 2026-04-04
URLs to Monitor
| URL | Label | Notes |
|---|---|---|
https://tensorix.ai/sitemap.xml |
Sitemap | Full site structure |
https://tensorix.ai/pricing |
Pricing | Pricing page |
https://tensorix.ai/models |
Models | Model catalogue |
https://docs.tensorix.ai/ |
Docs | Documentation hub |
https://tensorix.ai/trust |
Trust | Compliance and security |
https://tensorix.ai/dedicated-inference |
Dedicated Inference | Enterprise offering |
https://tensorix.ai/blog |
Blog | Announcements |
https://tensorix.ai/alternatives/openai |
vs OpenAI | Competitive positioning |
Strategy
- Privacy-as-moat: Tensorix leads with zero data retention and EU sovereignty as primary differentiators, targeting the growing segment of EU enterprises re-evaluating US cloud dependencies
- Open-source model focus: Exclusively serves open-source models (now including OpenAI's OSS releases), positioning against vendor lock-in from proprietary providers
- Drop-in migration: Emphasises one-line SDK migration from OpenAI, reducing switching costs
- Compliance-forward: Publicly available DPA, sub-processor list, and SLA — unusual transparency for the space
- NVIDIA Inception member: Part of NVIDIA's startup programme, suggesting GPU/infrastructure partnership
- Content marketing: Active blog comparing against OpenAI and Anthropic, publishing EU AI Act guides, positioning as thought leader on sovereign AI
Formidability
Score: 4/10
Tensorix occupies a real niche (EU-sovereign inference) and has significantly expanded its model catalogue from ~8 to 14 confirmed LLMs including Llama 4 Maverick (1M+ context), OpenAI OSS models, and Qwen 3. No visible observability features and limited documentation remain weaknesses. The EU sovereignty angle is compelling for a specific segment but restricts TAM. They compete more directly with OpenAI's API than with OpenRouter — they route to open-source models only, while OpenRouter offers both proprietary and open-source across many more providers. Low overlap with OpenRouter's core value prop (unified access to all models/providers), but worth monitoring as EU data sovereignty demand grows and their model catalogue expands.