IONOS AI Model Hub vs. Tabnine: data protection per plan

yes (data protection package of IONOS Cloud GmbH as DPA under Art. 28 GDPR, version 7.4, as of 06/2023; according to the documentation part of the contract without separate signature and without a product-specific annex for the AI Model Hub)

not publicly documented (Trust Center lists a Master Services Agreement, no DPA; Privacy Policy describes Tabnine as processor when used for organisations, but does not refer to any DPA)

not publicly documented (Trust Center lists a Master Services Agreement, no DPA; Privacy Policy describes Tabnine as processor when used for organisations, but does not refer to any DPA)

Terms of use for AI services section 2.3: input and output data only for providing the service under the contract, technical security and troubleshooting (improvement of the services only for SaaS); any further use, 'in particular for training, retraining or otherwise developing AI models, is expressly prohibited' (translated). Documentation 'Data Handling': prompts and outputs are 'not used to train, fine-tune or otherwise improve any model' and are not disclosed to the model developers.

Docs 'Privacy': 'no-train-no-retain' policy for Tabnine models; Tabnine does not store code, does not share code with third parties and does not train on customer code. Tabnine's code completion and Protected Chat model are stated to be trained only on open-source code under permissive licences; private fine-tuned models (team opt-in) accessible only to the team. Optional third-party chat models: privacy rules and protection may differ (Docs 'AI Models'). Terms of Use (as of 2024-07-25): Tabnine uses code developed during use only to improve the service for that specific customer, not for other users; "Tailor Made" models only at express request.

Docs 'Privacy': 'no-train-no-retain' policy for Tabnine models; Tabnine does not store code, does not share code with third parties and does not train on customer code. Tabnine's code completion and Protected Chat model are stated to be trained only on open-source code under permissive licences; private fine-tuned models (team opt-in) accessible only to the team. Optional third-party chat models: privacy rules and protection may differ (Docs 'AI Models'). Docs: even with private deployment, code in the context window remains ephemeral; in self-hosted operation, clusters send metrics and logs to Tabnine servers, in air-gapped operation they do not. According to the Docs, no code and no PII are sent in the process. Terms of Use (as of 2024-07-25): Tabnine uses code developed during use only to improve the service for that specific customer, not for other users; "Tailor Made" models only at express request.

Documentation 'Data Handling': 'All inference is performed in IONOS CLOUD data centers in Germany'; access to operational data only by personnel in the EU, no access from outside the EU for operation, maintenance, support or fault diagnosis. Product page: '100% GDPR-compliant and securely hosted in Germany' (translated; provider statement). The model weights come from third-party developers who, according to the documentation, have no access to customer data.

Security documentation: SaaS on AWS (North Virginia, USA) and Google Cloud (Council Bluffs, Iowa, USA), plus Heroku (Virginia, USA) as main locations. Docs 'Deployment options': Secure SaaS is 'hosted, managed, and monitored by Tabnine', without stating a location. The Trust Center names Google Cloud Platform under 'Infrastructure'. An EU region or choice of region is not documented.

Operation and location are the customer's responsibility (data centre or supported cloud provider). Telemetry: in self-hosted operation, metrics, logs and client telemetry go to Tabnine servers, location not documented.

• IONOS SE (Germany): technical support, support and operation of the AI Model Hub and of data centres (sub-contractor list, as of 10/2026)
• Documentation 'Data Handling': 'No third-party sub-processors are engaged for AI Model Hub'; model developers only supply the model weights and do not process personal data on the customer's behalf

• Hosting providers according to the security documentation: Amazon Web Services (main data centre North Virginia, USA), Google Cloud Platform (Council Bluffs, Iowa, USA), Heroku (Virginia, USA)
• Google Cloud Identity (identity management according to the security documentation)
• Optional third-party chat models (depending on selection): Anthropic Claude, OpenAI GPT, Google Gemini and others; according to the documentation, privacy rules of these models may differ
• According to the Privacy Policy, also payment service providers, analytics and marketing service providers as well as affiliated companies (e.g. Tabnine Inc.), without names
• Complete sub-processor list: not publicly documented (tabnine.com/subprocessors still returns HTTP 502 when retrieved on 2026-10-18; the Trust Center contains no sub-processor list)

• No operation by Tabnine; optionally connected third-party models via private endpoints at the customer's choice
• Telemetry recipient: Tabnine servers (hosting as for SaaS according to the security documentation: AWS, GCP, Heroku in the USA)

Usage-based, net prices excl. VAT, offer for businesses only (price list IONOS Cloud GmbH, as of 2026-10-18), per 1 million input/output tokens e.g.: Qwen3.5-9B EUR 0.10/0.15; Llama 3.1 8B Instruct EUR 0.15/0.15; Mistral Nemo Instruct EUR 0.15/0.15; Mistral Small 24B Instruct EUR 0.10/0.30; gpt-oss-120b EUR 0.15/0.65; Llama 3.3 70B Instruct EUR 0.65/0.65; Qwen3.8-27 EUR 0.40/2.70; Qwen3.5-397B-A17B EUR 0.60/3.60; Qwen3-Coder-Next EUR 0.15/0.80; LightOnOCR 2 EUR 0.15/0.30. Images: FLUX.1 [schnell] EUR 0.0288 per image; FLUX.2 [klein] 4B EUR 0.013 per first megapixel. Embeddings from EUR 0.01 per 1 million tokens.

not publicly documented (tabnine.com/pricing redirects to the Tricentis contact form; Terms: fees in USD, excluding taxes)

not publicly documented (tabnine.com/pricing redirects to the Tricentis contact form)

yes (inference API for language, code, OCR, image, embedding and reranking models from third-party developers, operated on IONOS infrastructure; OpenAI-compatible API, tool calling, RAG)

yes (code completion, chat, agent, CLI, context engine; Tabnine's own models as well as optional third-party models)

yes (as SaaS; models run in the customer's own data centre or VPC, private endpoints for further models)