OpenCode provides free access to DeepSeek V4.1 Flash with a 200K context limit.

CoinDesk reports:

IT Home, October 15: OpenCode has just announced that DeepSeek V4.1 Flash is now available for free, with a tested context window of 200K, down from the 1M context window version.

This free access is an additional step built upon the earlier "Operation Cheepseek" initiative. In September, OpenCode temporarily increased the usage quota for DeepSeek V4.1 Flash in the OpenCode Go plan from $15 to $60, and then adjusted it to $80 on October 12. With this new free access, calls to this model on the OpenCode platform no longer consume any quota.

DeepSeek V4.1 Flash was officially released on September 10, 2026, and is a native multimodal MoE model with a backbone of 552B parameters and an additional 196B Engram conditional memory parameters. IT Home notes that its architecture employs a causal encoder-decoder design, with the first 20 of 40 language layers serving as causal encoders and the latter 20 as decoders, reducing prefill computation by nearly half. The model does not offer a free API tier on the official DeepSeek platform; billing is conducted via prepaid pay-as-you-go.

Regarding user feedback, generation speed (TPS) has been a focal point of community discussions. Several users have reported significant fluctuations in speed. One user noted a peak output speed of approximately 490 tokens/s on the Ollama cloud platform, but overall performance has been inconsistent, with speeds dropping as low as 50–60 tokens/s during midday to afternoon hours.

Within the OpenCode Go plan, users have also reported TPS dropping below 100. Additionally, community testing shows that performance decreases as context length increases, with users reporting TPS falling below 100 when context exceeds 100K, and dropping to approximately 50 tokens/s at 200K.