Learned image compression with joint autoregressive context and hyperprior has achieved excellent performance. To avoid the time-consuming serial decoding pipeline, the checkerboard context model (CCM) with fast two-pass coding is proposed. However, CCM sets half of the latents as anchors to extract spatial context, which is rough and redundant. We propose a more precise and flexible content adaptive checkerboard context model to decrease the numbers and bit consumption of anchors. By introducing pseudo-anchors for simple regions in latents, our method can preserve the fast two-pass coding and outperform CCM in Rate-Distortion performance on several baseline models with negligible computational overhead.