However, DeepSeek moved its flagship V4 Pro from preview to general availability. DeepSeek V4 Pro Takes Aim at AI Giants With Near-Frontier Performance at a Fraction of the Cost.
Furthermore, the upgrade is positioned as a low-cost challenger to leading AI models such as Anthropic’s Claude Fable 5.
DeepSeek V4 Pro Takes Aim at AI Giants With Near-Frontier Performance at a Fraction of the Cost BEIJING: Chinese artificial intelligence company DeepSeek has moved the flagship V4 Pro model from preview to general availability. The new version is available through the company’s app, website and API. Its model weights are released under the MIT license. DeepSeek says the upgrade brings a major improvement in AI‑agent and coding performance.
DeepSeek’s benchmark results show sharp gains over the April preview. This includes a jump in Terminal-Bench 2.1 from 72.1 to 87.9. CyberGym rose from 52.7 to 83.3. Additionally, independent testing has not validated these figures.
The biggest attraction is cost. Additionally, Claude Fable 5 leads by a few percentage points across several agent benchmarks. Meanwhile, its token costs are higher. DeepSeek’s pricing remains dramatically below Claude Fable 5. The company has raised V4 Pro’s API rates compared with its earlier preview pricing.
Moreover, the launch highlights an increasingly competitive AI market. Additionally, Chinese developers seek frontier-level performance while keeping access costs low. This trend could pressure established US AI firms.
