Reasoning Models
Reasoning models are a new class of AI that spends extra compute at inference time to “think before answering” — producing dramatically better results on complex maths, code, and multi-step problems while being slower and more expensive on simple tasks. Understanding when this trade-off is worth it is now a core builder skill.
In This Section
What is Test-Time Compute
The core concept behind reasoning models — how spending more compute at inference (not training) unlocks fundamentally better reasoning.
OpenAI o-Series (o1, o3, o4)
Deep dive on OpenAI's reasoning model family — o1, o3, o4-mini, and how they compare for coding, maths, and agentic tasks.
DeepSeek-R1 & Open Reasoning
How DeepSeek proved frontier reasoning is achievable open-weight — the training approach, distillation, and what it means for self-hosted deployments.
When Reasoning Helps (and When Not)
A practical decision guide — which tasks benefit from reasoning models, which are hurt by extra latency/cost, and how to route intelligently.