OpenAI’s faster reasoning model
OpenAI released a new reasoning model that reportedly solves math and coding problems about 40% faster on select benchmarks, promising speed gains for analytical and coding workflows. (x.com)
OpenAI announced GPT‑5.4 mini and GPT‑5.4 nano on March 17, 2026 as smaller, faster variants of its GPT‑5.4 family aimed at coding, tool use, and high‑volume agent workloads. (openai.com: ) The company says GPT‑5.4 mini runs “more than 2x faster” than the previous GPT‑5 mini and that its latency estimates account for tool‑call duration, sampled tokens and input tokens. (openai.com: ) OpenAI’s published benchmark snapshots show GPT‑5.4 family strong on agentic and coding tests — the flagship hit an OSWorld‑Verified score of 75.0% and GDPval of 83% in launch materials referenced by reviewers. (buildfastwithai.com: ) OpenAI says GPT‑5.4 mini is available in ChatGPT, the Codex app and the API, while GPT‑5.4 nano ships to the API as the lowest‑cost, lowest‑latency option. (openai.com: ) Developers flagged that the new mini and nano tiers carry higher per‑token prices versus older “mini” generations — community notes and pricing roundups show input/token cost increases of roughly 3× for mini and up to 4× for nano in some comparisons. (LetsDataScience: ) Early adopter signals included immediate integrations and tests: OpenAI cited partners testing mini in production workflows, and reviewers reported companies like GitHub moving mini into Copilot‑adjacent tooling on launch day while independent benchmarkers continue to validate provider claims. (openai.com: )
