Inkling 2 Small: Open-Weight Model Claims Western Lead, Token Costs Rise

What happened

Thinking Machines, the startup led by former OpenAI chief technology officer Mira Murati, is releasing Inkling 2 Small, a mixture-of-experts model with 594 billion parameters. The company says it is the strongest western open-weight model available, a claim reported by The Decoder.

The release is aimed at agentic performance, meaning the model's ability to carry out multi-step tasks rather than simply answer a single prompt. The reporting notes a trade-off: stronger agentic results come with a sharp rise in token usage, which directly affects how much each task costs to run.

Why it matters

Open-weight models give organisations more control than closed, API-only services. Teams can host, inspect and adapt a model within their own infrastructure, which matters for data governance, customisation and long-term planning. A claim of leading performance in that category will draw attention from enterprises and public bodies weighing their options, though independent benchmarking will be the real test of that claim.

The token trade-off deserves as much attention as the headline capability. For digital transformation leaders, the unit that counts is the cost and speed of completing a task for a customer or employee, not the score on a leaderboard. Models that think through more steps can deliver better outcomes, but only if the operating model accounts for the extra compute, latency and spend that come with them.

The Renascence take

Most coverage will focus on whether Inkling 2 Small beats its rivals. The more useful question is what a given improvement costs per completed journey, and whether the customer notices the difference.

A model that performs better on agentic tasks is only an experience upgrade if the answer arrives quickly, costs the business a sustainable amount, and resolves the customer's problem in fewer loops. Treat token consumption as a service-design variable, not a back-office detail. Pilot the model on one high-volume journey, measure time-to-resolution and cost per outcome, and only then decide whether the extra capability earns its place.