OpenAI has made a dramatic move in the AI pricing wars while simultaneously stirring major controversy with its Pentagon partnership. On March 17, 2026, the company unveiled GPT-5.4 mini and nano models, with the nano priced at just $0.20 per million input tokens—beating even Google's Gemini 3.1 Flash-Lite at $0.25.
Ultra-Cheap AI Pricing Shakes the Market
The new gpt-5.4-nano represents OpenAI's most aggressive pricing yet. At $0.20 per million input tokens (with cached input at just $0.02), the model is now the cheapest on the market. For context, Claude Opus 4.6 charges $5.00 per million input tokens—25 times more expensive. The nano model can process 76,000 photos for just $52, making it viable for massive image analysis tasks.
According to OpenAI's self-reported benchmarks, GPT-5.4-nano outperforms the previous GPT-5 mini model when run at maximum reasoning effort. The new mini model is also 2x faster than its predecessor. Full pricing tiers include: GPT-5.4 at $2.50 input/$15 output, GPT-5.4-mini at $0.75 input/$4.50 output, and GPT-5.4-nano at $0.20 input/$1.25 output per million tokens.
Codex Subagents: Multi-Agent Revolution
Also launched this week, Codex Subagents represents OpenAI's push into parallel custom agents designed for complex tasks. This marks a notable advancement in multi-agent AI systems, enabling more sophisticated and parallel processing of intricate workflows. The technology allows multiple specialized agents to work simultaneously on complex problems, a significant step beyond single-model approaches.
Pentagon Partnership Sparks Outrage
In a development raising serious ethical questions, OpenAI has controversially agreed to provide the Pentagon access to its AI technology. Defense officials revealed the technology could assist in selecting strike targets—marking the first time generative AI advice is being actively applied to combat operations, reportedly being tested in Iran.
The partnership with Anduril, which manufactures drones and counter-drone technologies, provides additional context for potential military applications. "There's pressure to integrate it quickly with existing military tools," one defense official told MIT Technology Review. The development has sparked concerns among OpenAI's employees and customers about acceptable use cases.
Grok Faces CSAM Lawsuit
Meanwhile, xAI's Grok is facing its own controversy. The company has been sued over AI-generated child sexual abuse material, with victims alleging that Grok was specifically designed to create pornographic content from photos of real people. This lawsuit highlights the booming market for custom deepfake porn and the urgent need for AI safety measures.
Anthropic is responding to such concerns by recruiting weapons experts with experience in "chemical weapons and/or explosives defense" to prevent "catastrophic misuse" of its AI systems.
What Comes Next
OpenAI's dual-track week—aggressive market expansion with cutting-edge pricing and concerning military applications—underscores the company's conflicting priorities. The GPT-5.4 nano's rock-bottom pricing could accelerate AI adoption across industries, while the Pentagon partnership may alienate safety-conscious researchers and users. Watch for competitor responses to the pricing war and continued fallout from military AI applications.