DeepSeek V3.2

DeepSeek V3.2 is the general-purpose counterpart to R1 — a conventional model built for breadth rather than deliberation, and priced aggressively enough to have moved the market's expectations of what capability should cost.

Its practical importance is as a price anchor. Capability that was premium-tier eighteen months ago is now available at a fraction of the cost, and that shapes what is worth building.

Cost-to-capability is the headline and it is real. For ordinary production work — drafting, summarising, extraction, code generation, structured output — it delivers quality comparable to models several times the price.

It handles code well for a general model, and the open-weights availability means self-hosting is an option if data residency matters more than convenience.

On the hardest reasoning it trails the leading flagships, and it is not the model to reach for when a wrong answer is expensive. Ecosystem maturity — integrations, tooling, documented prompt patterns — is behind the US labs.

For organisations with constraints on where models may be hosted or where data may be processed, the governance question needs answering before the capability question.

A strong default for cost-sensitive production work, and worth benchmarking against your current model on your actual tasks — the price difference is large enough that even a modest quality gap may be acceptable. Escalate to a flagship for the small fraction of work where it is not.