OpenCode 加码免费模型调用，DeepSeek V4

OpenCode adds free model calls, and DeepSeek V4.1Flash is open for use. OpenCode announced that DeepSeek V4.1Flash is now available for free on its platform. The free version has been measured to support 200K context windows, which is lower than the 1M context version previously provided by this model. This opening means that users calling this model within OpenCode will no longer consume package credits. This adjustment is a further advancement of OpenCode’s previous “Operation Cheepseek” plan. In September this year, OpenCode temporarily increased the usage limit of DeepSeek V4.1Flash in the OpenCode Go package from US$15 to US$60, and further increased it to US$80 on October 12. After this free opening, calls within the platform will no longer be included in the quota consumption. In contrast, the DeepSeek official platform currently does not provide a free API level and adopts a pre-recharge and 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 and 196B Engram conditional memory parameters. It adopts a causal encoder-decoder architecture. The first 20 layers of the 40-layer language backbone are causal encoders and the last 20 layers are decoders. The amount of Prefill calculation is nearly halved. However, there are still fluctuations in model generation speed. Community user feedback shows that its output speed varies significantly on different platforms and time periods; in some tests, after the context exceeded 100K, the generation speed dropped to less than 100 tokens per second, and when it reached 200K, it was about 50 tokens per second. via AI News (author: AI Base)