Inkling 2 Small now available on AI Gateway

Thinking Machines released Inkling 2 Small on Vercel AI Gateway, adding a multimodal reasoning model tailored for coding agents and professional workflows. According to Vercel Blog's report, the architecture processes text and image inputs while producing text outputs. The system executes extended programming sessions, gathers contextual data with tools, and synthesizes research into deliverables.

This technical briefing examines model capabilities, visual feedback mechanisms, unified gateway routing, and client integration patterns. It is written for software engineers, systems architects, and technical leads deploying autonomous agents into production environments. Readers will find exact configuration steps, copy-paste terminal commands, and architectural guidance for agentic systems.

01What shipped in Inkling 2 Small

Thinking Machines introduced Inkling 2 Small as a specialized multimodal reasoning engine for agentic software engineering and analytical knowledge work. The model ingests multimodal prompts containing text and image data to return structured text completions. Its runtime design sustains focus across multi-step development sessions, navigating complex codebases and complex logic chains without losing contextual coherence.

Beyond software development, the model functions as an automated research assistant capable of autonomous tool execution. Inkling 2 Small queries external utilities to collect operational context and compiles that raw information into structured office assets. Development teams can instruct the model to produce formatted technical documentation, detailed data spreadsheets, and comprehensive analytical deliverables directly from environment queries.

02What improved in Inkling 2 Small

Visual comprehension in Inkling 2 Small expands beyond basic image recognition into detailed UI inspection and visual development. The model reads typography embedded in images, parses infographics, and extracts quantitative trends from multi-axis data charts. Software engineers can provide full-page interface screenshots, allowing the reasoning core to critique visual alignment and front-end layouts.

This multimodal capability establishes an automated feedback loop for front-end programming tasks. During visual development, the model reviews rendered application screenshots to verify the visual accuracy of generated code against requirements. When visual regressions or styling mismatches appear, Inkling 2 Small diagnoses rendering bugs and guides the next iterative code revision.

03What you gain from Inkling 2 Small

Deploying the model through Vercel AI Gateway provides production reliability advantages over calling upstream model endpoints directly. The platform manages an intelligent routing layer equipped with automatic retries, active failover paths, and latency optimizations. These infrastructural protections ensure agent workflows sustain higher-than-provider uptime across high-concurrency enterprise pipelines.

Organizations also gain strict governance tools to audit operational expenditures and isolate private telemetry. AI Gateway provides built-in custom reporting, granular budget caps on API keys, Zero Data Retention compliance, and routing rules. Vercel passes provider pricing directly with zero markup, charging no additional platform fees even on Bring Your Own Key requests.

04How to get Inkling 2 Small

Developers integrate the model into agentic coding environments such as Claude Code, Codex, and Cursor using the official Vercel CLI. Executing the setup command configures local routing rules and links credentials before selecting the thinkingmachines/inkling-2-small model name. The command creates necessary local configuration files so coding agents authenticate cleanly against the gateway without manual token formatting.

npx vercel ai-gateway setup

/model thinkingmachines/inkling-2-small
claude --model thinkingmachines/inkling-2-small

Application builders can invoke the model through the Vercel AI SDK or standard Chat Completions endpoints using the thinkingmachines/inkling-2-small identifier. The AI SDK streamText method provides responsive token streaming for TypeScript functions, interactive assistants, and background agent loops. In addition to coding assistants, developers can test model prompts directly in the web-based playground interface.

```typescript
import { streamText } from 'ai';

05What to watch after Inkling 2 Small

Engineers evaluating the model can test multimodal prompt dynamics and visual parsing directly inside the web-based model playground. Vercel also provides Reference Architecture blueprints detailing structural decisions across agent stacks, web applications, and backend runtimes. These curated resources guide development teams in orchestrating durable workflows, model tool bindings, authentication schemas, and background tasks.

Teams can accelerate assistant configuration using Vercel's centralized prompt library and downloadable Markdown catalog. These ready-to-use prompts provide scaffolding for bootstrapping application prototypes, extending agent toolsets, and refining project code. As multimodal adoption matures, monitoring runtime token consumption and screenshot resolution requirements will help maintain optimal cost efficiency.