OpenAI has just released GPT-5.4, the newest iteration of its reasoning-focused thinking models, continuing the company’s push toward systems designed for deliberate multi-step problem solving rather than quick conversational responses. GPT-5.4 expands the capabilities introduced in earlier GPT-5 models with improvements to reasoning reliability, tool usage, and complex task execution.
A central change is the model’s stronger reasoning mode, which (for better or worse) allocates more compute to internal reasoning steps before producing an answer. GPT-5.4 also improves instruction following and reduces failure modes where earlier models would drift from the user’s request. The model shows stronger integration with external tools and coding workflows, allowing it to manage tasks like multi-file code analysis and reasoning more reliably.
Early leaderboard results suggest the update is already competitive with the strongest frontier systems. At the time of the writing of this Story, on our AIW Model Leaderboard, GPT-5.4 scored 0.573 placing second overall, only one millesimal point under the mighty Gemini 3.1 Pro, which does not understate its impressive features at all since it is outperforming Anthropic’s Claude Opus 4.6 and Claude Sonnet 4.6. Not to mention, within OpenAI’s own lineup, the release marks a clear improvement over their earlier models.
What cannot be ignored is that the launch lands in the middle of an escalating political dispute around military AI. The US administration recently ordered federal agencies to phase out Anthropic’s systems after the company refused to allow its models to be used for domestic surveillance or autonomous weapons, arguing those uses violated its safety policies [New York Post]. At the same time, OpenAI moved to replace Anthropic in several government deployments, signing agreements to provide its models to the U.S. defense ecosystem while insisting that its technology would not be intentionally used for domestic surveillance. The situation has sparked a growing online backlash, framed as the Cancel ChatGPT movement, criticising the company’s role in military AI and warning about the risk of large-scale surveillance systems.
Against that backdrop, GPT-5.4 arrives not just as a technical upgrade, but as another flashpoint in the debate over how frontier AI systems should be used - and who controls them.