Updates·October 16, 2026, 05:10

Free access to DeepSeek's new model on OpenCode, but with less memory

AI-generated from the sources listed, with automated source checks.

According to Chinese media, OpenCode has made DeepSeek V4.1 Flash free, but the free version can only remember 200,000 tokens at a time. The model is a cheaper replacement for DeepSeek's previous flagship model, V4-Pro.

The platform OpenCode announced on October 15, 2026, that DeepSeek's V4.1 Flash model is now free to use. That's according to a Chinese tech outlet citing IT Home.

The free version does come with a limitation. It has a context window of 200K tokens, while the paid version supports 1M tokens. A token is a small piece of text, and the context window determines how much the model can handle at once, such as a long conversation or a large amount of code. After the launch, calls to the model on OpenCode no longer count toward a user's quota. It is not clear whether the free offer is time-limited.

The run-up at OpenCode

The free offer follows the "Operation Cheepseek" initiative. In September, the quota in the OpenCode Go plan was raised from $15 to $60, and on October 12 it was adjusted to $80.

Users report large swings in speed. On some platforms, around 490 tokens per second is mentioned, while the figure at peak load is 50-60 tokens per second. With more than 100K of context, it drops to under 100 per second, and at 200K of context it is around 50. These are users' own observations and not measurements from OpenCode.

About the model itself

DeepSeek launched V4.1 Flash on September 10, 2026. According to The Next Web, the model has 552 billion parameters in total. Parameters are the settings an AI model learns during training, and they are often used as a rough measure of a model's size. Other tallies cite a different total size of around 748 billion including Engram memory.

The model supports up to 1 million tokens of context and has been released under an MIT license on Hugging Face. That means others can freely download and use it.

On DeepSeek's own API, off-peak output costs $0.60 per million tokens. Starting September 14, DeepSeek is routing all requests to the older V4-Pro on to V4.1 Flash at the Flash price.

How good is it?

According to The Next Web, DeepSeek's own benchmark figures, meaning standardized tests, put the model on par with Claude Opus 5 on some coding tests, but it trails on others, such as ProgramBench and Terminal-Bench 3.0. It also lags on difficult reasoning. The figures have not been independently verified, so they should be treated with caution.