What is the cheapest model on Cloudflare Workers AI for Atlas?

The cheapest model on Cloudflare Workers AI for Atlas is IBM Granite 4.0 H Micro, priced at $0.017 per Mtok for input. This makes it the most cost-effective option in the registry for tasks like generating titles and subagent summaries within Atlas.

How do Cloudflare Workers AI context windows compare to other providers?

Cloudflare Workers AI generally offers trimmed context windows compared to other providers. For example, Qwen3 30B A3B is 32,768 tokens, and Llama 3.3 70B fp8 Fast is 24,000 tokens. Other hosts can provide much larger contexts, such as 262,144 tokens for Kimi K2.7 Code or GLM-5.2.

Can I use Cloudflare Workers AI for large codebases in Atlas?

While Cloudflare Workers AI can be used with Atlas for large codebases, its trimmed context windows for some models might limit the depth of analysis for very extensive files or projects. For tasks requiring the broadest context, models with 262,144 tokens on other hosts might be more suitable.

What are the authentication requirements for Cloudflare Workers AI?

To use Cloudflare Workers AI with Atlas, you need to export both your `CLOUDFLARE_ACCOUNT_ID` and `CLOUDFLARE_API_KEY` environment variables. Authentication is scoped to your existing Cloudflare account, simplifying setup.

Is Cloudflare Workers AI suitable for all Atlas agentic tasks?

Cloudflare Workers AI is highly suitable for cost-sensitive and latency-critical Atlas agentic tasks, especially those leveraging its cheaper models like IBM Granite 4.0 H Micro. However, for tasks demanding the largest context windows or specific models at their absolute lowest market price, other providers might be a better fit.

How does Cloudflare Workers AI's edge network benefit Atlas?

Cloudflare Workers AI's edge network benefits Atlas by running inference close to the user. Requests terminate at the nearest edge Point of Presence, rather than a single region, which significantly reduces latency and provides faster responses for Atlas's interactive and agentic operations.

Are there any models that are significantly more expensive on Cloudflare Workers AI?

Yes, some models are significantly more expensive on Cloudflare Workers AI. For instance, GPT OSS 120B costs $0.35 per Mtok for input here, which is roughly 9 times more expensive than its $0.037 per Mtok input price on DeepInfra, despite being the same model weights.

Atlas with Cloudflare Workers AI in 2026

Cloudflare Workers AI offers Atlas developers a unique blend of cost-effectiveness and edge network performance in 2026, making it an excellent choice for specific tasks. Its IBM Granite 4.0 H Micro model provides the registry's single cheapest input price at $0.017 per Mtok, making it ideal for `small_model` roles like generating titles and subagent summaries. While some context windows are trimmed compared to other hosts, its fast inference on Cloudflare's global edge network ensures low latency for many agentic operations within Atlas.

What is Cloudflare Workers AI best for in Atlas?

Cloudflare Workers AI excels as a cost-effective solution for specific Atlas tasks in 2026, particularly leveraging IBM Granite 4.0 H Micro at just $0.017 per Mtok. This makes it the registry's cheapest input token, perfect for minimizing overhead costs.

For Atlas users in 2026, Cloudflare Workers AI is best utilized for roles where cost efficiency and rapid inference are paramount. The IBM Granite 4.0 H Micro model, priced at an industry-leading $0.017 per Mtok for input and $0.112 per Mtok for output, is an unparalleled choice for the `small_model` slot in Atlas. This configuration allows Atlas to push the cost of generating titles, summarizing subagent activities, and handling other brief, frequent interactions to near zero. Additionally, GLM-4.7-Flash, available at $0.0605 per Mtok for input and $0.40 per Mtok for output, provides a capable reasoning model directly on Cloudflare's edge network. This ensures that even more complex agentic tasks benefit from low-latency execution, as requests terminate at the nearest edge Point of Presence rather than a single, distant region. The authentication process is also streamlined, scoped to a Cloudflare account that many developers may already possess, simplifying setup for Atlas.

What are the cost and context window tradeoffs with Cloudflare Workers AI?

While Cloudflare Workers AI offers compelling pricing for some models, developers in 2026 must consider its trimmed context windows and higher costs for certain larger models. For instance, Qwen3 30B A3B is limited to 32,768 tokens here.

Developers evaluating Cloudflare Workers AI for Atlas in 2026 should be aware of specific tradeoffs, particularly concerning context windows and pricing for certain models. The context windows on Cloudflare Workers AI are often trimmed relative to other hosts. For example, Qwen3 30B A3B offers 32,768 tokens, Llama 3.3 70B fp8 Fast is limited to 24,000 tokens, and QwQ 32B also provides 24,000 tokens. This can be a significant limitation for Atlas agents performing deep code analysis or working with very large files, where a broader context is crucial. Furthermore, while some models are exceptionally cheap, others can be more expensive. GPT OSS 120B, for instance, costs $0.35 per Mtok for input and $0.75 per Mtok for output on Cloudflare Workers AI. This is roughly 9 times more expensive on input compared to its pricing of $0.037 per Mtok for input and $0.17 per Mtok for output on DeepInfra, despite being the same model weights. This means that while Cloudflare Workers AI offers a curated, fast catalog, the cost-effectiveness can vary dramatically depending on the specific model chosen and its availability on alternative providers.

When should I choose a different model provider over Cloudflare Workers AI?

In 2026, you should consider alternative model providers if your Atlas workflows demand the largest available context windows or if you require specific models at their absolute lowest market price. Other hosts offer context windows up to 262,144 tokens.

While Cloudflare Workers AI is excellent for specific use cases within Atlas, there are clear scenarios where a different model provider would be more suitable. If your primary requirement is access to the largest possible context windows for extensive code analysis, complex refactoring, or processing vast amounts of documentation, Cloudflare Workers AI's trimmed context windows may be a limiting factor. Other providers offer models like Kimi K2.7 Code and GLM-5.2 with a substantial 262,144 token context window, or Nemotron 3 Super 120B and Gemma 4 26B A4B with 256,000 tokens. These larger windows are critical for Atlas's Axis, the hybrid semantic and keyword code search, or for agents drafting plans across an entire codebase. Additionally, if cost optimization for specific high-demand models, such as GPT OSS 120B, is your top priority, you might find more competitive pricing on other platforms. As noted, GPT OSS 120B is significantly more expensive on Cloudflare Workers AI compared to providers like DeepInfra. Cloudflare Workers AI's catalog is deliberately small, which can be a limitation if your Atlas agent requires a model not offered or if you need a broader selection of specialized models.

Setup

01Export your Cloudflare account ID and API key: Ensure both `CLOUDFLARE_ACCOUNT_ID` and `CLOUDFLARE_API_KEY` environment variables are set in your shell.
02Verify environment variables: Run `atlas models cloudflare-workers-ai` to confirm that Atlas successfully resolves both environment variables and can connect to the service.
03Configure the base URL: The base URL for Cloudflare Workers AI is templated with your account ID: `https://api.cloudflare.com/client/v4/accounts/${CLOUDFLARE_ACCOUNT_ID}/ai/v1`.
04Pin the agentic coding model: In your `atlas.json` configuration, set `"model": "cloudflare-workers-ai/@cf/moonshotai/kimi-k2.7-code"` for the primary agentic coding slot.
05Optimize for small model tasks: Set `"small_model": "cloudflare-workers-ai/@cf/ibm-granite/granite-4.0-h-micro"` in `atlas.json` to leverage the cheapest input tokens for titles, summaries, and subagent overhead.

Frequently asked questions

What is the cheapest model on Cloudflare Workers AI for Atlas?: The cheapest model on Cloudflare Workers AI for Atlas is IBM Granite 4.0 H Micro, priced at $0.017 per Mtok for input. This makes it the most cost-effective option in the registry for tasks like generating titles and subagent summaries within Atlas.
How do Cloudflare Workers AI context windows compare to other providers?: Cloudflare Workers AI generally offers trimmed context windows compared to other providers. For example, Qwen3 30B A3B is 32,768 tokens, and Llama 3.3 70B fp8 Fast is 24,000 tokens. Other hosts can provide much larger contexts, such as 262,144 tokens for Kimi K2.7 Code or GLM-5.2.
Can I use Cloudflare Workers AI for large codebases in Atlas?: While Cloudflare Workers AI can be used with Atlas for large codebases, its trimmed context windows for some models might limit the depth of analysis for very extensive files or projects. For tasks requiring the broadest context, models with 262,144 tokens on other hosts might be more suitable.
What are the authentication requirements for Cloudflare Workers AI?: To use Cloudflare Workers AI with Atlas, you need to export both your `CLOUDFLARE_ACCOUNT_ID` and `CLOUDFLARE_API_KEY` environment variables. Authentication is scoped to your existing Cloudflare account, simplifying setup.
Is Cloudflare Workers AI suitable for all Atlas agentic tasks?: Cloudflare Workers AI is highly suitable for cost-sensitive and latency-critical Atlas agentic tasks, especially those leveraging its cheaper models like IBM Granite 4.0 H Micro. However, for tasks demanding the largest context windows or specific models at their absolute lowest market price, other providers might be a better fit.
How does Cloudflare Workers AI's edge network benefit Atlas?: Cloudflare Workers AI's edge network benefits Atlas by running inference close to the user. Requests terminate at the nearest edge Point of Presence, rather than a single region, which significantly reduces latency and provides faster responses for Atlas's interactive and agentic operations.
Are there any models that are significantly more expensive on Cloudflare Workers AI?: Yes, some models are significantly more expensive on Cloudflare Workers AI. For instance, GPT OSS 120B costs $0.35 per Mtok for input here, which is roughly 9 times more expensive than its $0.037 per Mtok input price on DeepInfra, despite being the same model weights.

Try SeaShell in your terminal

The terminal-native AI coding agent. Free core, single binary.

Install SeaShell

Related guides

Atlas vs Cosine: Choosing Your AI Coding Agent in 2026

Compare Atlas and Cosine AI coding agents for 2026. Atlas offers terminal-native TUI, explicit change review, and BYO model keys. Cosine features its Lumen models and a 30.08% SWE-bench record.

Atlas for dbt: Terminal-Native AI Coding in 2026

Atlas is a terminal-native AI coding agent for dbt. Read the ref() DAG, convert a table model to incremental, run dbt build against dev, and add tests in 2026.

Research a Third-Party API Before Integrating It with Atlas in 2026

How to research a third-party API with Atlas in 2026: websearch finds the current docs, webfetch pulls the page as markdown or text, and grep checks repo conventions.

Atlas vs Jules: Choosing Your AI Coding Agent in 2026

Comparing Atlas and Jules in 2026: Atlas offers terminal-native TUI and local execution, while Jules provides cloud-based async GitHub integration with a Critic agent.

Atlas vs Traycer: Terminal AI Coding Agents in 2026

Compare Atlas, a terminal-native AI coding agent, with Traycer, an orchestration layer for existing agents, in 2026. Evaluate features, pricing, and workflow.

Atlas vs Sourcery: Terminal AI Coding Agents in 2026

Comparing Atlas, the terminal-native AI coding agent, with Sourcery, the Python-first AI reviewer, in 2026. Evaluate features, pricing, and workflow for developers.

Migrate a Deprecated API Across Every Callsite with Atlas (2026 Workflow)

How to migrate a deprecated API across every callsite with Atlas in 2026: the lsp tool's findReferences enumerates callers, todowrite tracks them, apply_patch migrates each one.

Atlas for Electron: Terminal-Native AI Coding for Main, Preload, and Renderer in 2026

Atlas is a terminal-native AI coding agent for Electron in 2026, where the main and renderer split, contextIsolation, and preload bridges are the security model.

Browse this resource hub