OpenCode announced that DeepSeek V4.1Flash is now available for free on its platform. The free version supports 200K context windows, which is less than the 1M context version previously offered by the model. This release means that users no longer need to use their plan when calling the model within OpenCode.
This adjustment is a further step in OpenCode’s previous “Operation Cheepseek” program. In September this year, OpenCode temporarily increased the usage quota for DeepSeek V4.1Flash in the OpenCode Go package from $15 to $60, and on October 12, it was further raised to $80. After this free access, calls within the platform are no longer counted toward the quota consumption. In contrast, the official DeepSeek platform currently does not offer a free API tier and uses a prepaid, pay-as-you-go model.
DeepSeek V4.1Flash was released on September 10, 2026. It is a native multi-modal MoE model with a backbone parameter size of 552B, along with 196B Engram conditional memory parameters. It uses a causal encoder-decoder architecture, where the first 20 layers of the 40-layer language backbone are causal encoders and the last 20 layers are decoders, resulting in nearly half the Prefill computational load.
However, there are still fluctuations in the model generation speed. User feedback from the community indicates that the output speed varies significantly across different platforms and time periods. In some tests, when the context exceeds 100K, the generation speed drops below 100 tokens per second, and at 200K, it is approximately 50 tokens per second.