Why Isn't the Industry Freaking Out About DeepSeek 4.1 Flash?

DeepSeek 4.1 Flash is starting to look like the kind of model that quietly rewrites habits. The writer says they’ve used it heavily for about a month across a dozen projects, and in the middle of a session they often can’t tell whether they’re working with DeepSeek or Opus. For day-to-day coding, planning, and even research, it behaves like a frontier model without carrying frontier-model pricing.

That matters because the economics are absurdly lopsided. With an OpenCode Go subscription that costs $10 a month, DeepSeek is described as basically unlimited. Small tasks can cost $0.003 instead of $1, and the writer says they’ve rarely gone over $1 in expected costs during a session, even when those sessions run all day.

The appeal isn’t just cheaper autocomplete. The writer leans on 4.1 Flash for complex planning and research, then uses Opus 5.5 only for occasional final code review before sending DeepSeek back in to make the fixes. The point is less about one model being perfect and more about having a cheap, capable engine for the bulk of the work.

A big part of that comes down to cache efficiency. DeepSeek is said to have shrunk the KV cache by roughly 437x versus its V1 model, which cuts one of the biggest costs in long coding sessions. The writer argues that’s why all-day use can stay under a dollar, and why the model’s economics make self-hosting a bad deal if the goal is saving money.

The broader claim is blunt: Chinese model makers are playing a different game, and the rest of the industry hasn’t reacted enough. The writer thinks the gap in capability is small enough now that most developers will choose value over pedigree, and that frontier labs built around expensive training runs are going to have to answer for that.