DeepSeek releases its V4 Pro model with competitive pricing compared to Claude Fable, which leads in benchmark tests.
DeepSeek Launches Updated Model
DeepSeek has announced the release of its V4 Pro model, now marked as the 0813 build, following a preview that began in April. This launch was noted without a formal announcement on the API pricing page, which reflects unchanged usage costs of $0.435 per million input tokens and $0.87 per million output tokens.
The latest benchmarks reveal that Claude Fable surpasses DeepSeek's performance by an average of 5.3% across nine tests, despite DeepSeek winning two of those tests. It’s important to note that the benchmarks prior to this release were based on a preview version of DeepSeek.
Pricing Comparison Highlights
DeepSeek’s pricing structure remains substantially lower than that of competitors, offering its services at roughly 98% less than GPT-5 Pro. Meanwhile, Claude Fable's pricing stands at $10 per million input tokens and $50 for output tokens, resulting in a cost comparison that shows V4 Pro as approximately 46 times cheaper than Fable 5.
When evaluating costs per task, the difference becomes more pronounced. For instance, completing a benchmark task costs $3.15 using Fable 5, while V4 Pro charges just 3 cents. This highlights a considerable advantage for DeepSeek in terms of affordability for businesses relying on AI tools.
Competitive Landscape and Future Developments
As competition heats up, Anthropic's Claude Opus 5 is noted for outperforming Fable 5 while maintaining a lower price point. DeepSeek has conducted its own internal evaluations but has not yet disclosed the infrastructure used for testing. Additionally, the company plans to release a new minimal mode for its model soon.
Amidst these developments, Chinese AI labs are also gaining ground, achieving competitive results at significantly reduced costs. For example, Kimi K3 has shown superior performance compared to both Fable 5 and GPT-5.6, illustrating a trend where expense reductions are becoming increasingly important in the AI landscape.
