DeepSeek just dropped the preview of their V4 series, with both open-weight and available via API. 1M context window. V4-Pro (full power, 1.6T params) and V4-Flash (smaller, faster, cheaper, 284B params).
API is already updated: https://api-docs.deepseek.com/