API Platform Access

Description

Provision of access to multiple artificial intelligence models through a single API platform, enabling teams to access, test and use a broad range of third-party AI models without needing to establish separate integrations or arrangements with individual model providers. The service provides a centralised interface for accessing models from multiple independent providers, alongside consistent account-level security, data and usage controls. Access to a broad range of frontier and open-weight AI models through a single API and account. Ability to access models from multiple independent model providers and deployments. Controls to restrict model providers based on agreed security, data protection and assurance requirements. Account-level controls for data retention and the use of customer data for model training. Access to newly released models and capabilities as they become available through the platform. Usage and spend monitoring and reporting to support financial management and invoice reconciliation. Appropriate service availability, maintenance and operational support arrangements. Technical and account management support for users. Ability to monitor model availability, usage and service performance across providers.

Key facts

Procedure

Direct award

Lots

1

Contact

commercialexcellence@ics.gov.uk

Contract period

1 Dec 2026 → 30 Nov 2027

Value incl. VAT

£1.2m

Category

services

Awarded suppliers

• OpenRouter Inc · GB-PPON-PLGJ-3815-XWXP · US, US

Ground given for limiting competition

The buyer’s own stated reason for awarding without open competition, as published. Their assertion, not a finding.

AISI intends to award the contract directly to OpenRouter under Section 41(1) of the Procurement Act 2023, relying on the direct award justification in Schedule 5, paragraph 6 (Single supplier – technical reasons). AISI considers that the technical requirements of the contract can currently only be met in full by OpenRouter, due to the combination of capabilities required and the absence of a reasonable alternative supplier capable of delivering the complete service. AISI requires a single, integrated platform providing fast, self-service access to a broad and continually evolving range of frontier and open-weight AI models through one API, contract and centralised control layer. The essential technical requirements include: Access to multiple independent inference and model-hosting providers through a single integration. Rapid availability of newly released and emerging AI models. The ability to access alternative hosts and providers to support comparative testing, performance evaluation and resilience. Consolidated billing, usage monitoring and centralised account controls. Consistent security, data governance and contractual requirements across model and inference providers. OpenRouter's platform provides this combination by aggregating multiple independent model and inference providers through a single managed service. This allows AISI to access and compare models across different providers without having to establish and maintain separate technical integrations, commercial arrangements and control environments for each provider. AISI has considered alternative approaches, including contracting directly with individual model providers, procuring dedicated hosting and self-hosting models. These alternatives can provide some of the required functionality but do not, based on AISI's assessment of the market, provide the full combination of model breadth, speed of availability, access to multiple independent hosting providers and consolidated controls required through a single integration. Procuring these capabilities separately from multiple suppliers would not provide an equivalent technical solution. It would require AISI to establish and maintain separate integrations, contracts, security assurance arrangements, billing processes and operational controls. This would fragment the required service and introduce additional technical and operational complexity. Self-hosting would also require significant additional engineering resources to deploy, optimise, secure and maintain models and the associated infrastructure.