# Thinking Machines Open Sources Inkling, a Multimodal AI Model Focused on Affordability and Censorship Resistance

The new open-source model offers customizable, enterprise-ready AI with unique features.

- **Source:** [VentureBeat (feeds.feedburner.com)](https://brifing.si/source/c51474dc-6b09-49dd-aaca-085f4ed00c84)
- **Published:** 2026-07-15
- **Topic:** [Data & AI](https://brifing.si/c/data-ai)
- **Original article:** <https://venturebeat.com/technology/thinking-machines-open-sources-first-multimodal-language-model-inkling-focused-on-low-cost-and-resistance-to-censorship>
- **On Brifing:** https://brifing.si/s/ad3797a1-03d7-464a-bfc0-650f094d93b3

## Introduction

Thinking Machines has open-sourced Inkling, a multimodal language model designed for customization and on-premise deployment. Licensed under Apache 2.0, Inkling stands out with its low-cost operation and resistance to censorship, offering enterprises a trustworthy AI solution.

## Technical Specifications

Inkling is a 975-billion-parameter Mixture-of-Experts (MoE) system capable of processing text, images, and audio. Available on Hugging Face and the company's Tinker API, the model emphasizes a balanced approach to performance and cost efficiency.

## Performance Benchmarks

Inkling achieves 77.6% on SWE-bench Verified and 91.4% on VoiceBench, outperforming Nvidia's Nemotron 3. However, Chinese models like GLM 5.2 and DeepSeek V4 Pro show superior performance in certain specialized tasks.

## Multimodal Capabilities

As one of the few open-source models handling multiple modalities, Inkling scores 73.3% on MMMU Pro (vision) and 77.2% on MMAU (audio). These results position it competitively against proprietary alternatives in multimodal tasks.

## Censorship Resistance

A key differentiator is Inkling's design to provide factual outputs even on sensitive topics. Thinking Machines emphasizes this makes it a more reliable option for enterprises concerned about content moderation biases in other models.

## Cost Control Mechanism

Inkling introduces a novel 'controllable thinking effort' feature, allowing developers to programmatically adjust the AI's reasoning intensity (0.2-0.99). This provides granular control over the performance-cost tradeoff in production environments.

## Inkling-Small Variant

The company also unveiled Inkling-Small, a 276-billion-parameter version optimized for low-latency use cases. This lightweight alternative caters to workloads where rapid response times and cost efficiency are critical.

## Competitive Landscape

Entering the crowded 2026 open-weight market, Inkling competes with China's GLM 5.2 while outperforming US rival Nemotron 3. Compared to closed models like GPT 5.6 Sol, it trails in peak performance but offers unique advantages.

## Use Cases

Inkling is particularly suited for enterprises needing customizable, on-premise AI with strong data privacy. Its multimodal nature supports diverse applications from content analysis to interactive systems requiring visual and auditory understanding.

## Strategic Vision

Thinking Machines departs from the 'god model' approach, instead prioritizing real-world adaptability. Inkling represents a shift toward practical, controllable AI systems that balance performance with operational costs.

## Conclusion

Inkling fills a crucial gap in the open-source AI ecosystem by combining affordability, multimodal capabilities, and censorship resistance. Its enterprise-friendly features position it as a compelling alternative to both proprietary systems and specialized open-weight models.

## Sources covering this story

- https://venturebeat.com/technology/thinking-machines-open-sources-first-multimodal-language-model-inkling-focused-on-low-cost-and-resistance-to-censorship — 

---

_Summary written by Brifing from the original reporting. Copyright in the original article belongs to its publisher; read the full piece at the link above._