This page is a projection of the one entry record, the Reliability and Data control factors that Implement covers. The full verdict is set by all four factors together, floor-weighted so the weakest caps the whole.
API integration
Infercom exposes an OpenAI-compatible API at https://api.infercom.ai/v1, so an existing
OpenAI client integrates by changing the base URL and API key. The docs also list Anthropic-SDK
compatibility and a Responses API. Infercom is reachable via the Opper AI gateway for teams
already routing through a gateway.
Region & sovereignty configuration
Region is chosen by model id, and the DPA covers only the EU ones. Models tagged "EU" in
/v1/models run in Munich. Non-EU models run on SambaNova's global infrastructure. DPA Sec
2a(4) obliges the customer to use only EU-hosted models unless a separate addendum covers
Non-EU Models.
There is no region switch and no ZDR toggle to manage: zero content retention is the platform
behaviour. Private connectivity is not offered ("Not today"), so the API is reached over public
HTTPS.
Rate limits, tiers & quotas
Limits are per model and counted in requests. The free tier (no payment method) allows 20
requests per minute and 20 per day. The developer tier (card linked) allows, for example,
gpt-oss-120b 150 per minute and 50,000 per day and gemma-4-31B-it 80 and 20,000. Enterprise
defaults are gpt-oss-120b 500 per minute and 100,000 per day, with custom limits on request.
Rate-limit headers are returned on every response.
Portability & exit
The exit path is clean and includes shipped single-tenant and air-gapped options. Infercom
speaks the OpenAI-compatible API over portable open-weight models, so the same checkpoints run
on GPU providers or self-hosted stacks. Dedicated Capacity reserves single-tenant racks in
Munich, and on-premises SambaRack deployment (air-gap capable, about 90 days from contract to
production, bring-your-own-checkpoint) removes the hosted dependency entirely.
The SambaNova RDU is a provider-side hardware dependency in hosted mode only. On termination
the DPA requires return or deletion of personal data, and account data is deleted within 30
days.
The ownership factors this domain covers, drawn from the one entry record.
3
ReliabilityDoes it stay up and stay secure?
ModerateISO 27001:2022 is certified and a public status page shows one major incident in under a year of operation, but pay-as-you-go has no SLA and the Reserved and Dedicated SLAs are not public.
How this scores (AOI sub-dimensions)
Reliability3/5whether it stays up, with an SLA and status historyA public Atlassian Statuspage (ev-status) has run since 2025-12-31 with per-model components, an incident feed, a scheduled-maintenance notice and one documented major incident (gemma-4-31B-it, 2026-06-19, resolved in about 21 hours).
Security4/5the controls protecting your traffic and dataISO 27001:2022 certificate (ev-iso), plus binding technical measures.
Compliance4/5which independent certifications and attestations it holdsISO 27001:2022 certificate LU-IS-20250253 from Proks Certification GmbH, valid 2025-12-16 to 2028-12-15, scope Inference as a Service (ev-iso, third-party attestation).
4
Doesn't extract your dataDo the binding terms keep your data and IP yours?
StrongOn the data-extraction axis your data stays yours: zero content retention, no training, fine-tuning or service improvement, metadata-only 90-day logs and customer_retains, per Privacy Statement v2.4, Terms v2.1 and DPA v1.4. The caveats are jurisdiction and contractual scope, not extraction: Global Catalogue prompts go outside the EEA (primarily US) and outside the DPA, and a US operator has remote administrative access.
How this scores (AOI sub-dimensions)
Data governance4/5retention, training-on-inputs and data ownershipZero content retention: "Infercom does not store, log, or retain the content of your prompts or model outputs".
Residency4/5where your data is processed and storedEU-hosted models run on Infercom-owned hardware at Equinix Munich 4 and, under DPA v1.4 (ev9, Sec 1, 2(1), 7a), are processed exclusively in the EU/EEA with no fallback or overflow routing outside it.
The same evidence records as the entry sheet. Read means the text was verified; unverified means it is known to exist but not yet read.
Vendor announcementread2026-11-19
Pricing page and home page: OpenAI-compatible API on SambaNova RDU hardware in Munich, with per-token EUR rates (gpt-oss-120b 0.22/0.59, Gemma 4 31B 0.20/0.35, Whisper 0.10, E5-Mistral 0.13, MiniMax M3.1 Flash 0.60/2.40 per 1M tokens, excluding VAT) and no minimums.
Vendor announcementread2026-11-19
Trust Centre (first-party surface): no persistent storage of prompts or outputs, and usage logs hold metadata only for 90 days.
Third-party analysisunverified2026-07-25
SambaNova partnership announcement: Infercom as a sovereign AI provider, Luxembourg HQ, EU jurisdiction, Munich deployment, ISO 27001, GDPR-safe/EU-compliant.
Third-party analysisread2026-11-19
SambaNova and Infercom Q&A: SambaNova partners with Infercom to launch a European sovereign inference cloud with all data under European jurisdiction.
Third-party analysisread2026-11-19
Independent gateway partnership page (updated October 2026): Infercom models are routed through the Opper AI gateway, OpenAI-compatible, EU German datacentres by default plus US and JP regions, native BF16.
Vendor announcementread2026-11-19
Zero Data Retention glossary: ZDR means content is not persisted after the request and is not used for training.
Vendor announcementread2026-11-19
EU Sovereign AI page and docs: EU sovereign models run in Munich; "Our US operating partner SambaNova has remote access to run the platform; the DPA names it"; "Luxembourg company, controlled in Europe, no US parent".
Privacy Policyread2026-11-19
Privacy Statement v2.4, last updated 2026-10-30 (binding): "Infercom does not store, log, or retain the content of your prompts or model outputs".
Data Processing Addendumread2026-11-19
Data Processing Agreement v1.4, 24 September 2026 (binding): an Art.
Privacy Policyread2026-11-19
Privacy Statement v2.4 Sec 8 and 8.1 (binding): for Global Model Catalog models not hosted on EU infrastructure, "your API requests (including prompt content) are routed to SambaNova's global infrastructure outside the EEA, primarily in the United States".
Documentationread2026-11-19
On-Premises page, Trust Centre FAQ and Sovereign AI page: shipped today are SambaRack SN40L on-premises racks, "Air-gapped deployment support", "Typically ~90 days from contract to production", BYOC ("fine-tunes of supported architectures load immediately") and one contract with Infercom delivered with SambaNova engineers.
Terms of serviceread2026-11-19
Terms v2.1 and Privacy Statement v2.4 give functional protections only (transient processing and discard, no secondary use, logical isolation, encryption, access controls) and NO express confidentiality clause over customer inputs.
Terms of serviceread2026-11-19
Terms of Service v2.1, 2026-03-09 (binding).
Securityread2026-11-19
ISO/IEC 27001:2022 certificate No LU-IS-20250253 issued by Proks Certification GmbH: certified company INFERCOM SCS, scope "AI & Cloud Infrastructure Services (Inference as a Service) and the systems that support these services", valid 2025-12-16 to 2028-12-15, IAF and DAkkS accreditation marks.
Securityread2026-11-19
CSA STAR registry listing "Infercom Inference-as-a-Service": STAR for AI Level 1, AI CAIQ self-assessment v1.1.0, listed 2026-07-02.
Slaread2026-11-19
Public status page (Atlassian Statuspage JSON API): Infercom Inference Service component since 2025-12-31, per-model components, one public incident (2026-06-19 to 2026-06-20, impact major, gemma-4-31B-it), and a scheduled maintenance notice for 2026-11-20.
Documentationread2026-11-19
Developer docs list the Inference API catalogue (EU-hosted gpt-oss-120b, gemma-4-31B-it, E5-Mistral-7B, Whisper-Large-v3; Global MiniMax-M3.1-Flash).