OpenAI on Wednesday unveiled GPT-5.4, an upgrade to its flagship model that adds a one-million-token context window and a built-in "agentic" mode capable of autonomously executing multi-step workflows across desktop and web software environments.
The company reported that GPT-5.4 scored 75 per cent on the OSWorld-V benchmark, a test simulating real productivity tasks such as filling spreadsheets, drafting emails and operating browser-based business applications. Earlier OpenAI models scored well below 50 per cent on the same benchmark.
GPT-5.4 will be made available to ChatGPT Plus, Team and Enterprise customers in the United States and Europe immediately, with API access for developers rolling out over the next several weeks. Pricing per token is unchanged from GPT-5.0.
The release coincides with reports that OpenAI has surpassed $25 billion in annualised revenue and is taking early steps toward a public listing, potentially as soon as late 2026. The company declined to comment on IPO timing.
Separately, OpenAI and data-warehouse company Snowflake announced a $200 million strategic partnership to integrate the model maker's most advanced systems directly into Snowflake's Data Cloud. The deal will allow joint customers to run autonomous agents over proprietary data without exporting it from the warehouse.
Independent benchmarking groups said they would publish evaluations within days, with particular interest in long-context reasoning and the rate of harmful or hallucinated outputs at the new context size.