On 16 April 2026, Cloudflare announced a significant expansion of its AI platform, positioning itself as a unified inference layer for AI models. The move, detailed in a blog post, aims to address the growing complexity of building AI agents that often require multiple models from different providers. For business leaders navigating AI adoption, this development signals a shift toward more integrated and manageable AI infrastructure.
The Problem: Model Sprawl and Agent Complexity
As AI models evolve rapidly, the best model for a given task can change within months. Cloudflare notes that the optimal model for agentic coding today might be replaced by a different model from another provider in three months. Moreover, real-world use cases often require calling more than one model. For example, a customer support agent might use a fast, cheap model to classify a user’s message, a large reasoning model to plan actions, and a lightweight model to execute individual tasks.
Refer & earn: 30% of a client’s first month, then 10% of every month after that, for up to 24 months per client. See the partner programme →
This multi-model approach creates operational challenges: managing costs across providers, ensuring reliability when one provider has an outage, and controlling latency for users in different locations. These challenges intensify when building agents, which can chain ten inference calls together for a single task. A slow provider doesn’t just add 50ms; it adds 500ms. A failed request can cascade into downstream failures.
Cloudflare’s Solution: One API for All Models
Cloudflare’s answer is to become a unified inference layer. Starting immediately, developers can call third-party models using the same AI.run() binding they already use for Workers AI. For those using Workers, switching from a Cloudflare-hosted model to one from OpenAI, Anthropic, or any other provider is a one-line change. For non-Workers users, REST API support is promised in the coming weeks.
Cloudflare now offers access to 70+ models across 12+ providers through one API, one line of code to switch between them, and one set of credits to pay for them. The model catalog includes open-source models hosted on Cloudflare Workers AI and proprietary models from major providers. Cloudflare is expanding access to models from Alibaba Cloud, AssemblyAI, Bytedance, Google, InWorld, MiniMax, OpenAI, Pixverse, Recraft, Runway, and Vidu, who will provide their models through AI Gateway. Notably, the catalog now includes image, video, and speech models, enabling multimodal applications.
Cost Management and Customization
Accessing all models through one API also centralizes AI spend management. Cloudflare cites that most companies today are calling an average of 3.5 models across multiple providers, which means no single provider can give a holistic view of AI usage. With AI Gateway, businesses get one centralized place to monitor and manage AI spend. By including custom metadata with requests, they can break down costs by attributes like free vs. paid users, individual customers, or specific workflows.
For companies needing custom models, Cloudflare is working on letting users bring their own model to Workers AI. The company notes that the overwhelming majority of its traffic comes from dedicated instances for Enterprise customers running custom models. To extend this, Cloudflare leverages Replicate’s Cog technology to help containerize machine learning models. Cog simplifies packaging by abstracting away CUDA dependencies, Python versions, and weight loading. Users write dependencies in a cog.yaml file and inference code in a Python file, then run cog build to create a container image and push it to Workers AI. Cloudflare is developing customer-facing APIs and wrangler commands for pushing containers, as well as faster cold starts through GPU snapshotting. They are testing internally and with external design partners.
Performance and Reliability for Agents
For live agents, speed is critical. Cloudflare emphasizes that a user’s perception of speed hinges on time to first token, not total response time. Even if total inference is 3 seconds, getting the first token 50ms faster can make an agent feel zippy rather than sluggish. Cloudflare’s network of data centers in 330 cities positions AI Gateway close to both users and inference endpoints, minimizing network time before streaming begins. Workers AI hosts open-source models on its public catalog, including large models purpose-built for agents like Kimi K2.5 and real-time voice models. When calling these Cloudflare-hosted models through AI Gateway, there’s no extra hop over the public Internet, reducing latency.
Reliability is equally important. Every step in an agent workflow depends on the previous steps, so one call failing can affect the entire downstream chain. Through AI Gateway, if a model is available on multiple providers and one goes down, Cloudflare automatically routes to another available provider without requiring failover logic from the developer. For long-running agents using Agents SDK, streaming inference calls are resilient to disconnects. AI Gateway buffers streaming responses independently of the agent’s lifetime. If an agent is interrupted mid-inference, it can reconnect to AI Gateway and retrieve the response without making a new inference call or paying twice for the same output tokens. Combined with the Agents SDK’s built-in checkpointing, the end user never notices.
Replicate Integration
The Replicate team has officially joined Cloudflare’s AI Platform team. Cloudflare states they no longer consider themselves separate teams. They are working on integrations, including bringing all Replicate models onto AI Gateway and replatforming hosted models onto Cloudflare infrastructure. Soon, users will be able to access models they loved on Replicate through AI Gateway and host models they deployed on Replicate on Workers AI.
What This Means for Businesses
For business leaders, Cloudflare’s unified inference layer offers a way to simplify AI operations. Instead of managing multiple provider relationships, billing, and failover logic, companies can use one API to access a broad catalog of models. This can reduce operational overhead, improve cost visibility, and enhance reliability for agent-based applications. The ability to bring custom models and the integration with Replicate further extends flexibility.
However, businesses should consider potential vendor lock-in. While Cloudflare provides a unified interface, relying on a single platform for all AI inference may create dependencies. Additionally, the promised REST API support and customer-facing tools for custom models are still in development, so not all features are immediately available.
As AI agents become more prevalent, infrastructure that supports multi-model workflows, cost management, and reliability will be crucial. Cloudflare’s move reflects a broader trend toward consolidation and simplification in the AI stack. Businesses evaluating their AI strategy should monitor these developments and assess how unified inference layers can support their goals.
Disclosure: Faida Workforce provides AI workforce training and consulting services. This article is for informational purposes and does not constitute an endorsement of Cloudflare.
Sources
This article was produced with the assistance of AI tools. A human editor is accountable for its content.
30% of their first month, then 10% every month after that
Refer a business to Faida Workforce and you earn 30% of the client’s first paid month, then 10% of every month for up to 24 months per client — rising to 15% from the accrual after you pass 10 active clients. Paid by PayPal, batched on the 1st.
10 Pro clients = $2,598.70 over 24 months — $3,748.13 at the 15% tier.
Figures are potential earnings worked from the published rates, assuming every referred client pays for the whole cap — not a promise, not an average. The $1 trial is never commissionable.
