GPT-5.4 is OpenAI’s newest model, and it’s the first one we’ve used that feels built for agent workflows: planning cleanly, delegating well, and consistently following through without getting lost halfway. That matters for how we build and ship in Augment, so we’ve made GPT-5.4 our new default model.
For a limited time, it’s also free to use in Augment, across Intent, the Auggie CLI, or our IDE extensions. Your existing threads won’t change unless you choose to switch models.
GPT 5.4 is free for a limited time
5.4 is built for this moment in AI
Agent orchestration is the new bottleneck. Model capability has made execution cheap. Agents can generate a lot of code quickly, but coordination is where workflows still collapse: specs drift, tool state gets lost, parallel efforts conflict, and the system slowly diverges from “what we meant” to “what happened.” The next leap isn’t raw codegen. It’s models that can manage long-horizon work: keep a plan coherent, update it when constraints change, and use tools and intermediate artifacts predictably.
GPT-5.4 feels tuned for that reality. In our testing it stays anchored longer, makes fewer “start over” turns, and behaves more like a reliable orchestrator. This is exactly the behavior you want inside Intent, where coordination is the product. But it’s not a specialist model you only reach for on big, multi-agent jobs. It’s a strong daily driver because the same traits that make orchestration work, like token efficiency on complex tasks and reliable tool use, also make day-to-day development smoother.
“Since launching GPT-5, we've focused on training models that can take on longer-horizon coding work. GPT-5.4 feels like the first model to cross the threshold on intelligence, communicativeness, and reliability, so developers can ship more with confidence in an agentic environment like Intent, where Augment's context engine helps keep the model anchored in the right code and context.”
What’s improved
Better token efficiency on hard tasks. On harder tasks like complex refactors, multi-file reasoning, and architectural planning, GPT‑5.4 uses ~18–20% fewer tokens in our internal testing. Fewer tokens means more work done for the same budget, and less “restart the whole approach” overhead on long runs.
More reliable tool use. GPT‑5.4 does a better job tracking where it is in a workflow, recovering from small mistakes, and completing multi-step sequences without repeating work.
Stronger orchestration. Multi-agent orchestration is one of the biggest step-ups. GPT‑5.4 is better at breaking a spec into clear tasks, respecting inter-agent dependencies, and keeping agents aligned on the same definition of done.
Try it now
Open Augment, pick GPT‑5.4, and try it on something real:
- Run a refactor.
- Ask for an architectural plan with clear steps.
- Spin up an Intent workflow and let it orchestrate the work end to end.
GPT‑5.4 is free to try in Augment for a limited time. Sign up today and start building.
Written by
John Edstrom
Director of Engineering
John is a seasoned engineering leader currently redefining how engineering teams work with complex codebases in his role as Director of Engineering at Augment Code. With deep expertise in scaling developer tools and infrastructure, John previously held leadership roles at Patreon, Instagram, and Facebook, where he focused on developer productivity and platform engineering. He holds advanced degrees in computer science and brings a blend of technical leadership and product vision to his writing and work in the engineering community.