DeepSeek launches V4-Pro model that rivals Kimi K3 on some benchmarks at $0.44 input and $0.87 output per million tokens

DeepSeek introduced V4-Pro as its newest and most advanced language model. The announcement positions the model as a direct competitor to Kimi K3, claiming parity on certain benchmark tests. This launch marks a new entry in the competitive landscape of large language models.

The pricing structure highlighted in the release sets V4-Pro at $0.44 per million input tokens and $0.87 per million output tokens. These rates are described as much lower than those of competing offerings. For developers and operators, the lower per-token cost could reduce the expense of running inference workloads.

Performance claims are limited to "some benchmarks" where V4-Pro reportedly rivals Kimi K3. The material does not state that the model matches or exceeds Kimi K3 across all evaluation suites. Consequently, its advantage may be task-specific and not universally applicable.

While the lower price and benchmark parity suggest a cost-effective alternative, the announcement provides no detail on other factors such as latency, memory footprint, or support for specialized features. Teams evaluating V4-Pro would need to assess whether its performance on the relevant benchmarks aligns with their specific use-case requirements.