OPENAI GPT-5.4 SHIPS: 1.05M CONTEXT, BUILT-IN COMPUTER USE, PRO TIER
OpenAI released GPT-5.4, a unified frontier model that combines reasoning, coding, and computer-use with a 1.05M-token context and an optional Pro tier for heav...
OpenAI released GPT-5.4, a unified frontier model that combines reasoning, coding, and computer-use with a 1.05M-token context and an optional Pro tier for heavier workloads.
In ChatGPT, the family appears as GPT-5.4 Thinking and GPT-5.4 Pro, while the API exposes gpt-5.4 and gpt-5.4-pro with configurable reasoning effort; Codex now defaults to this line, consolidating coding and reasoning in one family (structure and specs, announcement). Pricing published via Puter’s API lists $2.5 per million input tokens and $15 per million output tokens, with up to 1,050,000 context and 128K max output model card and pricing.
Computer-use is now built in: the model can drive desktop UIs and websites and posted 75% on OSWorld-Verified, exceeding a human baseline; it also set new highs on BrowseComp and GDPval, indicating stronger research and professional task performance (feature and benchmarks, capabilities). For vision and document scenarios, OpenAI’s cookbook details prompt and workflow tips to maximize accuracy and throughput with the latest model developer tips.
Teams can automate full end‑to‑end workflows that span code, documents, web, and desktop actions in one model.
The 1.05M context window reduces chunking and orchestration overhead for large repos, logs, and data packs.
-
terminal
Evaluate computer‑use for repetitive back‑office tasks (browser/data entry) and gated admin actions using custom confirmation policies.
-
terminal
Load real codebases and large specs into the 1M‑token context and measure retrieval accuracy, latency, and cost.
Legacy codebase integration strategies...
- 01.
Plan migration from gpt‑5.2 and gpt‑5.3‑codex to gpt‑5.4/5.4‑pro and re‑baseline unit/integration tests for coding agents.
- 02.
Update tool routing and token budgets; validate permissions and sandboxing for UI automation in staging.
Fresh architecture paradigms...
- 01.
Design agents around built‑in computer‑use and reasoning.effort controls to trade off latency vs. accuracy.
- 02.
Use the large context to simplify retrieval and avoid complex chunking pipelines in new services.