Salvo Software

DeepSeek V4 just took the price lead. 15-30x cheaper than Claude and GPT-5.

Fortune reported that if Uber had used DeepSeek for their AI operations instead of Claude, the budget that ran out in four months would have lasted eight years.

Read that again.

DeepSeek V4 Flash sits at $0.14 per million input tokens through their cloud API. Claude Haiku is around $1. Same chat-style API, same kind of integration, very different invoice. Kimi K2.6, the other Chinese frontier model that landed last month, matches GPT-5.4 on coding tasks at 75% less cost.

When AI costs drop this dramatically, the math on what you can ship changes. Apps too expensive to run continuously start running 24/7. Startups can compete with enterprise budgets. The cost-per-token line item turns into a strategic variable.

What AI applications would you build if costs dropped 90% overnight? Tell us in the comments below.

OpenAI just launched GPT-5.4-Cyber, expanding their Trusted Access program to vetted cybersecurity defenders. This isn't just another AI model release. It's a signal that AI is becoming the primary battlefield for cyber defense, not just an auxiliary tool.

The implications for businesses are massive. While enterprises scramble to adopt AI for productivity, threat actors are already weaponizing the same technology. The companies that survive the next decade will be those that build AI-native security architectures from day one, not bolt-on solutions.

What concerns me most? The "vetted defenders only" approach suggests the capabilities are so powerful that unrestricted access would be dangerous. That's both reassuring and terrifying.

Full details on OpenAI's expansion: https://openai.com/index/scaling-trusted-access-for-cyber-defense