GPT‑5.4 hits 1.05M context
Diego Michelato reports OpenAI’s GPT‑5.4 lineup now includes 4 variants with up to a 1.05 million token context window, an 83% GDPVal score and ~33% fewer errors compared with previous models. (x.com)
OpenAI shipped GPT‑5.4 on March 5, 2026 and deployed it across ChatGPT (as “GPT‑5.4 Thinking”), the public API, and Codex. (openai.com)) The company merged the coding capabilities of GPT‑5.3‑Codex into the mainline GPT‑5.4 and added built‑in computer‑use that lets agents interact with software in a build‑run‑verify‑fix loop. (openai.com)) The release introduced a new tool_search feature that uses deferred tool loading to load only relevant tool definitions and improve tool‑selection accuracy in large tool ecosystems. (developers.openai.com)) Independent reviewers reported roughly a 47% reduction in token usage for tool‑heavy workflows when using GPT‑5.4’s tool_search optimizations. (digitalapplied.com)) OpenAI and third‑party tests show gains in computer‑use benchmarks, with OSWorld scoring GPT‑5.4 at about 75%, above the 72.4% human baseline cited in the reports. (dev.to)) API pricing lists GPT‑5.4 at $2.50 per 1M input tokens and $15.00 per 1M output tokens, and prompts with more than 272K input tokens are charged at 2x input and 1.5x output for the full session. (developers.openai.com)) OpenAI also launched GPT‑5.4 mini and nano as lower‑latency, cost‑efficient variants for coding and sub‑agent workloads; mini is available in ChatGPT while nano is delivered via the API. (openai.com)) Developers can tune internal effort via the Responses API using reasoning.effort (options include low/medium/high/xhigh), and OpenAI/community notes indicate most reported benchmark numbers were obtained with the xhigh setting. (developers.openai.com))
