Key Takeaways
- Thinking Machines Lab, founded by former OpenAI executive Mira Murati, has released its inaugural AI model named Inkling.
- Inkling is an open‑weight model that can be downloaded and customized by developers, though its training data and source code remain undisclosed.
- The company positions Inkling as a versatile, cost‑efficient system that processes text, image, audio, and video queries while balancing performance against expense.
- Benchmarks show Inkling competes well with other open‑weight offerings, though Thinking Machines admits it is not the strongest model currently available, closed or open.
- The startup aims to address a perceived gap in the U.S. market for competitive open AI models, especially as Chinese developers advance and major U.S. labs shift toward proprietary systems.
- While Inkling itself is free to use, Thinking Machines generates revenue from its developer tool Tinker, which helps customers fine‑tune models for specific tasks such as financial analysis.
- Murati envisions Inkling as part of a broader effort to build “interaction models” that capture the subtleties of human‑AI collaboration, including pauses, interruptions, and unspoken cues.
- The company raised $2 billion at a $12 billion valuation last year, but several early employees have since departed for firms like Meta and OpenAI.
Overview of Inkling’s Release
Thinking Machines Lab announced on Wednesday that it has unveiled its first artificial intelligence model, dubbed Inkling. The launch comes more than a year after Mira Murati, a former OpenAI executive, founded the startup in February 2023. According to the company’s statement, Inkling is designed to be “versatile and efficient, processing queries across different media while balancing ‘cost against performance.’” This positioning suggests the model aims to handle a mixture of text, image, audio, and video inputs without demanding prohibitive computational resources—a selling point for enterprises wary of spiraling AI expenses.
Open‑Weight Nature and Developer Access
A notable feature of Inkling is its open‑weight status. Thinking Machines emphasizes that developers can download the model and tailor it to their needs, even though they will not gain visibility into the underlying training data or source code. This approach mirrors the strategy of other open‑weight releases, such as Meta’s Llama family, while differentiating Inkling from fully open‑source projects where both code and data are shared. By keeping the training corpus private, the startup hopes to protect its intellectual property while still fostering a community of builders who can fine‑tune the model for niche applications.
Performance Claims and Benchmark Context
The company asserts that Inkling “was trained for strong performance across the board,” yet it candidly adds that “it is not the strongest model available today, closed or open.” Independent benchmarks cited in the announcement indicate that Inkling holds its own against comparable open‑weight alternatives, particularly in tasks requiring multimodal understanding. However, Thinking Machines does not claim superiority over leading closed models like GPT‑4 or Claude 3, acknowledging that the current frontier of AI performance remains dominated by proprietary systems. This modest framing may be intended to manage expectations while highlighting the model’s practical utility for cost‑conscious users.
Strategic Positioning in the AI Landscape
Thinking Machines Lab is part of a wave of so‑called AI neolabs that emerged after the exodus of talent from established labs such as OpenAI and Anthropic. These startups aim to push the frontier of AI software while offering alternatives to the dominant players. The release of Inkling is portrayed as an attempt to fill a perceived void in the United States market, where “Chinese developers” are seen as outpacing American firms in delivering competitive open AI products. As major U.S. players like Meta pivot toward closed, monetizable models and OpenAI’s portfolio remains heavily weighted toward paid offerings, Thinking Machines bets that developers and enterprises will gravitate toward freely available, adaptable tools like Inkling.
Funding, Valuation, and Personnel Shifts
The startup’s financial backing underscores investor confidence in its vision. Thinking Machines raised $2 billion at a $12 billion valuation last year, a figure that placed it among the most heavily funded AI ventures of the period. Despite this capital infusion, the company has experienced notable turnover; several early employees have departed for roles at Meta Platforms Inc. and OpenAI. Such movements reflect the highly competitive talent market in AI, where expertise frequently shifts between incumbents and emerging challengers. The exodus does not appear to have derailed the company’s product roadmap, as evidenced by the timely launch of Inkling.
Revenue Model: The Tinker Tool
While Inkling itself is offered at no direct cost, Thinking Machines generates revenue through its developer‑focused product Tinker. Tinker enables users to fine‑tune and customize AI models for specific workflows, a service the company sells to clients such as hedge fund Bridgewater Associates to enhance performance on financial tasks. This model mirrors the approach of other AI startups that monetize tooling and support rather than the base model itself, allowing Thinking Machines to sustain operations while promoting broad adoption of its open‑weight offering.
Vision for Interaction Models
Beyond raw processing power, Mira Murati has articulated a broader ambition for Inkling: to serve as a foundation for “interaction models” that facilitate more natural collaboration between humans and AI. In a recent interview with Bloomberg, she remarked, “Our interactions with each other are very rich. There’s a lot of information in our interactions – when we’re silent, when we’re thinking, when we’re interrupting one another. Interaction models are able to capture all of these nuances.” This statement underscores the belief that future AI systems must go beyond simple query‑response patterns and instead interpret the subtleties of human communication, such as pauses, tone, and turn‑taking, to become truly collaborative partners.
Implications for the U.S. AI Ecosystem
The introduction of Inkling arrives amid growing national concern over the cost and accessibility of cutting‑edge AI. As businesses scrutinize their AI expenditures, many are turning to freely available models from Chinese developers for certain workloads, prompting policymakers to weigh the strategic implications of relying on foreign technology. By providing an open‑weight alternative that balances cost and performance, Thinking Machines seeks to offer a domestically sourced option that could reduce dependence on overseas models while still satisfying performance requirements. Whether Inkling will shift the balance in the U.S. market remains to be seen, but its launch signals a clear intent to compete in the evolving landscape of accessible AI.
Conclusion: A Measured Step Forward
Thinking Machines Lab’s debut with Inkling reflects a calculated blend of ambition and pragmatism. The model’s open‑weight nature, multimodal competence, and focus on cost efficiency address immediate developer needs, while the startup’s broader vision of interaction models points toward a longer‑term goal of making AI a more intuitive collaborator. Backed by substantial funding and led by a high‑profile founder, Thinking Machines is positioned to influence the next phase of AI development—particularly if it can translate its technological promise into widespread adoption and sustainable revenue through tools like Tinker. Only time will tell whether Inkling becomes a cornerstone of the U.S. open‑AI ecosystem or a valuable niche player among many alternatives.
https://fortune.com/2026/07/15/what-is-mira-murati-thinking-machines-first-ai-model-inkling/