OVHcloud AI Endpoints
Listed at https://www.ovhcloud.com/en/public-cloud/ai-endpoints/
Overall Rank #41 ⭐ Consider
⚠️ Proxy required for mainland China. This is a European sovereign endpoint (Gravelines, France) with no mainland-China region; OVHcloud's own Batch comparison table lists availability as WW (out of US). Mainland production traffic needs a proxy or aggregator. | 🌍 International
💰 Token Pricing
| Type | Price | Note |
|---|---|---|
| Input | 按 1M token 计费，EUR 目录价：gpt-oss-20b 0.04、gpt-oss-120b 0.08、Qwen3-Coder-30B 0.07、Qwen3.5-9B 0.10、Mistral-Small-3.2-24B 0.10、Mistral-7B 0.11、Qwen3.8-27B/3.6-27B 0.40、Qwen3.5-397B-A17B 0.60、Llama-3.3-70B 0.67、Qwen2.5-VL-72B 0.91；嵌入 bge-m3 0.01、Qwen3-Embedding-8B 0.10 | per million tokens |
| Output | 按 1M token 计费，EUR 目录价：Qwen3.5-9B 0.15、gpt-oss-20b 0.15、Qwen3-Coder-30B 0.26、Mistral-Small-3.2-24B 0.31、gpt-oss-120b 0.40、Llama-3.3-70B 0.67、Qwen3.8-27B/3.6-27B 2.70、Qwen3.5-397B-A17B 3.60；输出单价约为输入价的 2.5-6 倍 | per million tokens |
💡 Free Credits: Free playground across all catalog models; anonymous tier of 2 req/min with no API key; Batch API free during the two-month Beta; Qwen3Guard, image generation and TTS models listed at 0
🤖 Supported Models (20)
gpt-oss-120b (reasoning LLM, 117B params, 131K ctx, EUR 0.08/0.40; USD 0.09/0.47)gpt-oss-20b (reasoning LLM, 21B params, 131K ctx, EUR 0.04/0.15; USD 0.05/0.18)Qwen3.5-397B-A17B (multimodal MoE, 262K ctx, EUR 0.60/3.60; USD 0.71/4.25)Qwen3.8-27B / Qwen3.6-27B (multimodal, 262K ctx, EUR 0.40/2.70; USD 0.47/3.19)Qwen3.5-9B (multimodal, 262K ctx, EUR 0.10/0.15; USD 0.12/0.18)Meta-Llama-3.3-70B-Instruct (131K ctx, EUR 0.67/0.67; USD 0.74/0.74)Qwen2.5-VL-72B-Instruct (visual, 32K ctx, EUR 0.91/0.91; USD 1.01/1.01)Qwen3-Coder-30B-A3B-Instruct (code, EUR 0.07/0.26)Mistral-Small-3.2-24B-Instruct-2506 (EUR 0.10/0.31; decommission 2026-11-04)Mistral-7B-Instruct-v0.3 (EUR 0.11/0.11)Qwen3-Embedding-8B (EUR 0.10 per 1M input), bge-m3 / bge-multilingual-gemma2 (EUR 0.01 per 1M input)whisper-large-v3 (EUR 0.00004083/sec; USD 0.163/hr), whisper-large-v3-turbo (EUR 0.00001278/sec; USD 0.051/hr)Qwen3Guard-Gen-8B / 0.6B (moderation, Beta, free); stable-diffusion-xl-base-v10 (image gen, free); nvr-tts en/it/de/es (TTS, free)
✨ Pros
- ✓The only major European cloud publishing per-1M-token inference rates in both EUR and USD on a public page
- ✓OpenAI-compatible API: migration means changing base_url and the model name; official OpenAI SDKs work unchanged
- ✓Batch API takes files up to 50k rows / 200 MB for asynchronous processing, with a documented 50% discount at GA and a free Beta
- ✓Data sovereignty: infrastructure in Gravelines, France, with the vendor stating data is not stored or shared during or after model use
- ✓Governed by Appendix 10, Specific Conditions for AI Endpoints, of the OVHcloud Public Cloud Special Conditions rather than a generic wrapper agreement
- ✓gpt-oss-120b at EUR 0.08/0.40 per 1M makes it one of the cheapest 100B-plus-class reasoning endpoints on this list
⚠️ Cons
- ×A narrow 20-model catalog with no frontier proprietary LLMs and no Anthropic models
- ×Vendor-declared availability is WW (out of US); US-resident inference is not satisfiable here
- ×API keys carry no spend or token ceiling, so a runaway loop has no built-in cap
- ×Models are retired on a public 3-month clock and then return 404; Mistral-Small-3.2-24B is already on the list while still appearing in the live price table
- ×Retirement notices go only to the account's billing contact, so the team actually running the API can miss them
- ×The playground caps LLMs at 1024 output tokens, making it unusable for judging long-generation quality
- ×Batch is still in Beta, the 50% discount applies at GA, and it carries no uptime SLA
🎯 Best For
European teams with an EU data-residency or procurement-compliance requirement currently paying a compliance cost for a US-hosted OpenAI-compatible endpoint; asynchronous batch workloads such as document classification, knowledge-base vectorisation and dataset preparation; teams that want LLM, embeddings, speech-to-text, moderation and TTS from one vendor with a single price list and a single contract appendix.