Use Inkling Small on Claude Code via AI Gateway

#ai #programming #tech #product

Inkling Small on AI Gateway gives Claude Code users a cost-effective coding agent with controllable thinking effort. Connect via CLI setup.

Key Takeaways

Inkling Small on AI Gateway gives Claude Code users a cost-effective coding agent with controllable thinking effort.
Connect via CLI setup.

What Changed — Inkling Small Lands on AI Gateway

Thinking Machines released Inkling Small, now available through Vercel's AI Gateway. The model delivers reasoning performance comparable to the full Inkling model at roughly one-quarter the size and compute cost. For Claude Code users who route through AI Gateway, this means a new, cheaper option for agentic coding, tool use, and visual tasks.

You can access it by setting model to thinkingmachines/inkling-small in the AI SDK. The model supports controllable thinking effort — you dial up reasoning depth when you need it, dial down when you want speed and lower cost.

What It Means For You — Cost-Efficient Agentic Coding

If you run Claude Code through AI Gateway (and you should if you want centralized logging, fallbacks, and zero data retention), Inkling Small becomes a strong secondary model. It's not a replacement for Claude's coding strengths, but it's a capable generalist that excels at:

Reasoning-heavy tasks where you'd normally pay for full-size models
Tool use and function calling — the model holds up well on agentic loops
Visual document analysis — Inkling Small can crop, zoom, and inspect images programmatically, useful for charts, diagrams, and screenshots in your codebase

The controllable thinking effort is the killer feature. You can set it to "minimal" for quick linting or formatting passes, and "maximum" for complex debugging sessions. This lets you optimize cost per task without switching models.

Try It Now — Connect Inkling Small to Claude Code

Set up AI Gateway for coding agents if you haven't already:

   vercel ai-gateway coding-agents setup

Configure your agent to use thinkingmachines/inkling-small as the model. In your Claude Code config or agent setup, specify:

   {
     "model": "thinkingmachines/inkling-small",
     "zeroDataRetention": true
   }

Test the thinking effort control by adding a parameter for reasoning depth. The exact API may vary, but the model supports a range from minimal to maximum reasoning.
For visual tasks (document screenshots, UI mockups, diagrams), Inkling Small's programmatic cropping and zooming means it can focus on small details — useful when you're asking about a specific error message or chart region.
Monitor costs via AI Gateway's dashboard. Since Gateway passes through provider pricing with no markup, you'll see exactly what you're spending per request.

Why This Matters for Claude Code Users

The Claude Code ecosystem is expanding beyond Anthropic's models. AI Gateway already supports multiple providers, and Inkling Small fills a specific niche: a smaller, cheaper reasoning model that's still good at agentic coding. If you're prototyping or iterating rapidly, it's a smart fallback or secondary model.

Zero Data Retention is also on by default if you toggle it — important for teams handling sensitive codebases. AI Gateway routes only to providers that delete prompts and responses after each request.

Looking Ahead

Inkling Small is still new, and its place in the Claude Code workflow will evolve. As Thinking Machines releases updates and the model matures, expect it to close the gap with full Inkling on complex agentic tasks. The open question: how will it compare to Claude's own small model options when Anthropic releases them? For now, Inkling Small is a solid, cost-effective choice that deserves a spot in your AI Gateway model rotation.

Source: vercel.com

[Updated 31 Jul via vercel_blog]

Inkling Small also natively reasons over audio, expanding beyond text and images for multimodal agentic tasks [per Vercel]. The model is available in the AI Gateway model playground for immediate testing, and supports Bring Your Own Key (BYOK) requests with no platform fee on inference. These details position it as a flexible, cost-efficient option for Claude Code users exploring multimodal workflows.

Originally published on gentic.news