Anthropic is prepping a hybrid reasoning model to challenge OpenAI

AI labs have spent the last few months splitting their lineups into normal fast chat models and slow-thinking reasoning engines. Anthropic seems ready to crush those two together.

According to a report from The Information, the startup is gearing up to release its next flagship model within weeks. Instead of forcing you to pick between speed and logic upfront, the upcoming release is described as a hybrid that can switch between deep reasoning and instant answers.

A sliding scale for your compute

Deep reasoning eats up expensive compute, so Anthropic will reportedly give developers a sliding scale to adjust how much thinking power—and money—they want to spend.

On performance, the report says the model beats OpenAI's o3-mini-high on several coding tasks, with notable strengths in parsing massive codebases and enterprise benchmarks.

Anthropic CEO Dario Amodei already dropped hints earlier in the week, saying he found the industry's strict divide between standard and reasoning models a little puzzling.

Why it matters

For developers, routing queries between lightweight chat endpoints and expensive reasoning engines has been a headache. A single model with adjustable depth eliminates that friction, especially for engineering teams analyzing sprawling software projects.

It also takes direct aim at OpenAI's product playbook. While OpenAI established two parallel tracks with GPT and the o-series, Anthropic is betting that an all-in-one hybrid is the true endgame.

We think the sliding scale approach looks smart on paper, but it puts the burden of budget optimization back on the user. If pricing isn't predictable, developers might still hesitate to slide that dial all the way up.