Skip to content

DeepSeek free: V4.1 Flash in OpenCode at no cost

DeepSeek free in OpenCode: V4.1 Flash with a 200K-token context no longer uses credits. What it can do, where the catches are and how to turn it on.

Piotr Sobolewski3 min read

This article in other languages:العربيةDeutschEspañolFrançaisItaliano日本語Polski简体中文

In this article
  1. Is DeepSeek really free?
  2. What can DeepSeek V4.1 Flash do?
  3. How to use DeepSeek for free in OpenCode
  4. What should we watch out for?
  5. FAQ

On October 15, the team behind OpenCode, the open-source agent harness that runs in your terminal, announced on X that DeepSeek V4.1 Flash is now free to use in it. So yes, DeepSeek free in a coding agent is a real thing right now.

Is DeepSeek really free?

A new model, deepseek-v4.1-flash-free, showed up in the OpenCode Zen documentation. According to Pandaily, calls to this model no longer use plan credits, and the free version has a 200K-token context instead of the full million.

Interestingly, this didn’t come out of nowhere. In September OpenCode raised the usage limit for this model in its Go plan from $15 to $60, and on October 12 to $80. Free access is the next step in the same direction.

The DeepSeek API itself still has no free tier, according to Pandaily, and the paid version of the model in OpenCode costs $0.30 per million input tokens and $1.20 per million output tokens (OpenCode Zen).

What can DeepSeek V4.1 Flash do?

DeepSeek released V4.1 Flash on September 10, 2026, with open weights under the MIT license. The key facts from the model card on Hugging Face:

  1. MoE (mixture-of-experts) with 552B parameters, of which only 8B are active per token during prompt processing and 16B during generation.
  2. Context of up to 1M tokens (200K in the free version in OpenCode).
  3. Multimodal: it takes text and images and answers in text.
  4. Adjustable reasoning: you set the effort level on a scale from 1 to 100.

According to DeepSeek’s own measurements (at maximum reasoning effort), the model scores 90.9 on GPQA Diamond, 90.6 on Terminal-Bench 2.1 and 74.2 on DeepSWE v1.1. These are the vendor’s numbers, though, so let’s take them with a grain of salt.

Its popularity, on the other hand, shows up in OpenCode’s own data: according to the stats page, V4.1 Flash was the number 1 model by token volume last week (62T), and 2.4 million people used it over two months.

How to use DeepSeek for free in OpenCode

  1. Install OpenCode and launch it in your project directory.
  2. Connect the OpenCode Zen provider following the instructions in the docs.
  3. Pick DeepSeek V4.1 Flash Free (deepseek-v4.1-flash-free) from the model list.

That’s it. From now on the agent runs on DeepSeek and we don’t pay for tokens. If we’re just getting started with terminal agents, most of the concepts carry over 1:1 from Claude Code.

What should we watch out for?

  1. Free “for now”. OpenCode says most of its free models are available “for a limited time”. Nobody knows how long the promotion will last.
  2. Speed drops as the context grows. Users quoted by KuCoin News report under 100 tokens per second above 100K tokens of context and about 50 tokens per second at 200K (these are community reports, not official measurements).
  3. Privacy. According to the OpenCode docs, the provider of this model has a zero-retention policy and doesn’t use data for training. Other free models on the same list may work differently, so it’s worth checking what we pick.

FAQ

Is DeepSeek free?

In OpenCode, yes: DeepSeek V4.1 Flash Free uses no credits and comes with a 200K-token context. The official DeepSeek API is still paid.

Does DeepSeek V4.1 Flash have open weights?

Yes, the weights are on Hugging Face under the MIT license.

How long will DeepSeek be free in OpenCode?

OpenCode hasn’t said. The free models in Zen are described as available for a limited time.