Recently, DeepSeek launched its upgraded model, DeepSeek V4, which features three selectable modes: Quick, Expert, and Visual. Most upgrades were initially focused on the web version, but changes have also been identified in the API model.
While the specific names of the models are not confirmed, analysis suggests that DeepSeek V4 Lite is the most likely option due to its smaller parameter size. A community member from Linux.do highlighted insights from internal beta testing, indicating that DeepSeek will offer two models with adjustable thinking modes, enhanced performance, and strong instruction adherence.
DeepSeek V4 Performance Compared to Competitors
Notably, the performance of DeepSeek V4 is reported to be close to that of Opus 4.5, although it struggles with more complex tasks compared to models like GPT. Despite Anthropic’s release of Opus 4.6 and 4.7, these models have faced criticism regarding performance and increased resource demands. This backlash has sparked interest in alternatives, such as DeepSeek V4.
Cost Considerations
- DeepSeek V4’s larger model likely to have increased costs relative to its predecessor, DeepSeek V3.X.
- Price anticipated to be lower than that of Opus models.
- Expected to maintain a favorable cost-performance ratio for developers.
As the performance of DeepSeek V4 improves, the pricing structure will likely adapt accordingly. While specific prices remain unannounced, it is expected that they will be higher than previous versions, but still more economical than competitors in the market.
Overall, if DeepSeek V4 can achieve performance levels comparable to Opus 4.5, it could emerge as a leading choice among open-source models, particularly for developers seeking robust yet cost-effective solutions.
