Fireworks launches DeepSeek V4 Flash with native vision at same price

What’s new

Fireworks announced that the DeepSeek V4 Flash model is now available with native vision through its platform. The new version replaces the text-only 0731 release and, according to the company, delivers better performance metrics while keeping the same pricing.

Pricing and availability

The model is offered in two tiers: serverless, a pay-as-you-go option without infrastructure management, and priority, which provides queue preference and guaranteed availability. The price is $0.22 per million input tokens and $0.66 per million output tokens (about 0.8 and 2.4 shekel respectively), identical to the previous version.

What native vision means

Native vision indicates that the model was trained from the start on both visual and textual data, rather than being assembled from separate language and vision models after the fact. This architecture could improve consistency across multimodal tasks and reduce computational overhead, but Fireworks has not released technical details about the training process.

Unverified performance claims

The company claims the new version outperforms its predecessor on performance metrics, but it did not disclose which benchmarks were used, the comparison baselines, or the sample size. Without published methodology or third-party verification, the claim remains a manufacturer statement.

How to try it

Developers can access the model now at fireworks.ai/models/deepsee... on the Fireworks platform.