We have a new weight class...

OpenAI has expanded its reasoning-focused o1 model family with the surprise launch of a new, smaller variant dubbed o1-mini. Announced via a brief update to its API documentation, the model is positioned as a more accessible entry point into OpenAI's 'reasoner' architecture. Priced at $0.15 per 1 million input tokens and $0.60 per 1 million output tokens, o1-mini is substantially cheaper than the flagship o1-preview ($15/$60) and o1 (reported at $7.50/$30), suggesting a deliberate strategy to create a new, lower-cost tier for the technology.

This move represents a significant shift in OpenAI's product strategy, introducing what the AI community is calling a new 'weight class.' While full performance benchmarks are not yet public, the model is expected to retain the core chain-of-thought reasoning capabilities of its larger siblings but at a reduced scale and cost. The release enables developers and companies to experiment with and integrate advanced reasoning AI into a wider range of applications, from complex data analysis to iterative coding tasks, where the expense of the larger o1 models was previously prohibitive.

The quiet rollout follows the viral success of the original o1 models, which distinguished themselves by 'thinking' through problems step-by-step before delivering an answer, leading to superior performance on reasoning-heavy benchmarks. By launching o1-mini, OpenAI is likely testing the market for a scaled-down reasoning engine and gathering data on usage patterns before a potential wider release or announcement. This tactic of stealthily adding models to the API is reminiscent of how other capabilities, like the Code Interpreter, were first introduced.