The $12 Billion Thinking Machines Startup Finally Shipped Inkling, a 975B Open-Weights AI Model That’s Deliberately Mid

The AI startup world has been buzzing for over a year about Thinking Machines, a $12 billion venture founded by a former OpenAI chief technology officer. After months of speculation about what the highly funded company would build, the team has finally released its first product: Inkling, a 975-billion parameter open-weights large language model (LLM) that the company explicitly positions as “deliberately mid” compared to top-tier frontier models like GPT-4o or Claude 3.5 Sonnet. The unusual positioning has sparked widespread debate about what the team is aiming to achieve, and why a startup with a $12 billion valuation would intentionally release a model that doesn’t compete at the very top of industry benchmark charts.

What Makes Inkling Different From Other AI Models?

Inkling is released under an open-weights license, which means developers and businesses can download the full set of model parameters, modify the code, fine-tune it for specific use cases, and run it on their own hardware without paying per-query API fees or adhering to strict usage caps set by a third-party vendor. This is a stark contrast to closed frontier models, which are only accessible via paid APIs, and often retain the right to use customer input data for future model training.

The “deliberately mid” positioning refers to the team’s choice to optimize Inkling for the most common real-world AI use cases, rather than chasing top scores on niche benchmarks for complex reasoning, advanced coding, or mathematical problem-solving. As the Thinking Machines team noted in their release announcement, 80% of enterprise AI deployments are used for routine tasks like customer support, data entry, content summarization, and basic workflow automation, not for solving complex research problems. Inkling is built to excel at those high-volume, everyday tasks without the bloat and cost of a frontier model.

Core Features of Inkling

• Commercial-friendly open license: Unlike many open-source LLMs that restrict commercial use for large deployments, Inkling’s license allows businesses of any size to use the model for commercial purposes without additional fees or legal restrictions.
• Pre-optimized for agent workflows: The model is fine-tuned out of the box to work with AI agents for common business tasks, reducing the need for extensive custom fine-tuning that other open-source models require for enterprise use.
• Low compute requirements: The Thinking Machines team optimized Inkling’s architecture to run efficiently on mid-tier GPUs, meaning small teams and startups don’t need to invest in hundreds of thousands of dollars worth of server hardware to deploy the model at scale.
• Built-in data privacy guardrails: Since the model can be run fully on-premises, sensitive customer data never leaves a business’s own infrastructure, making Inkling compliant with strict data regulations like HIPAA for healthcare and GDPR for European customer data.

Why Release a “Mid-Tier” Model At All?

The decision to skip the frontier model race is a deliberate strategic choice, not a sign that Thinking Machines can’t build a top-tier LLM. The team is targeting a massive underserved market: businesses that need affordable, compliant, easy-to-deploy AI, rather than the most capable model on the market.

A 2025 Gartner report found that 72% of enterprises cite high AI API costs as a top barrier to scaling their AI deployments, while 68% report concerns about data privacy and vendor lock-in with closed frontier models. Inkling directly addresses both of these pain points. For a business processing 100,000 customer support tickets per month, running Inkling on a $5,000 mid-tier GPU cluster costs a fraction of what they would pay for API access to a closed frontier model, with the added benefit of full control over their data.

The model also fills a gap in the open-source AI ecosystem. Existing open-source models like Llama 3 are highly capable, but require extensive fine-tuning and technical expertise to adapt for specific business use cases, and often have restrictive commercial licenses for large deployments. Mistral and other smaller open models are cheaper to run, but lack the parameter count and pre-optimization for complex agent workflows that Inkling offers.

Real-World Use Cases for Inkling

Inkling is already being tested by a range of enterprise and small business users for a variety of high-impact use cases:

• Customer support automation: Inkling can handle common customer queries, pull answers from internal knowledge bases, and escalate complex issues to human agents without additional fine-tuning, reducing support ticket resolution times by up to 60% in early tests.
• Internal workflow automation: Teams are using Inkling-powered agents to process invoices, sort and categorize incoming emails, update CRM records, and schedule meetings, cutting down on repetitive administrative work.
• Small business AI access: Unlike frontier models that are cost-prohibitive for small teams, Inkling’s low compute requirements make enterprise-grade AI accessible to local retail shops, independent agencies, and early-stage startups that previously couldn’t justify AI deployment costs.

What Inkling Means for the Future of AI

Thinking Machines’ decision to release a “deliberately mid” open-weights model is a signal that the AI industry is moving beyond the “bigger is better” arms race that has dominated the last two years. For most businesses, the most valuable AI tool is not the one that can solve complex math problems, but the one that is affordable, easy to deploy, compliant with data regulations, and built for the tasks they actually need to complete.

While frontier models will continue to push the boundaries of what AI can do, Inkling represents a pragmatic shift toward building AI tools that solve real-world problems for the majority of users, rather than just chasing benchmark headlines. For developers and businesses tired of high API costs and vendor lock-in, Inkling is a compelling new option worth exploring.

For a full breakdown of Inkling’s performance benchmarks, licensing terms, and step-by-step deployment walkthrough, watch the full video analysis linked below.