computer-use-preview and gpt-4o-search-preview
2025-03-11 · Publisher : OpenAI · Domain : Artificial Intelligence
OpenAI opens the Responses API with three specialized models for agents: computer-use-preview to operate a computer, gpt-4o-search-preview and gpt-4o-mini-search-preview for web search.
What changes
- computer-use-preview is based on Operator's CUA model and powers the computer use tool of the Responses API.
- The search models are versions of GPT-4o and GPT-4o mini trained to understand and run web queries, with citations, in Chat Completions.
- The launch comes with the Responses API, the file search tool and the Agents SDK.
- The vendor plans to align the Assistants API with Responses, with a shutdown announced for mid-2026.
### Specifications - computer-use-preview: text and image input, text output, Responses API only, research preview for developers in usage tiers 3 to 5. Snapshot computer-use-preview-2025-03-11; the documentation lists 8,192 tokens of context and 1,024 of output. - gpt-4o-search-preview and gpt-4o-mini-search-preview: 128,000-token context, 16,384-token output, text only, knowledge cutoff October 1, 2023. ### Pricing Pricing per million tokens (USD) and per tool call. | Model | Input | Output | Other | | --- | --- | --- | --- | | computer-use-preview | 3 | 12 | | | gpt-4o-search-preview | 2.50 | 10 | from $30 per 1,000 requests | | gpt-4o-mini-search-preview | 0.15 | 0.60 | from $25 per 1,000 requests | The per-token prices of the search models come from the current documentation; the announcement only gives the per-request price. ### Benchmarks Scores published by the vendor. | Benchmark | Result | | --- | --- | | OSWorld (computer use) | 38.1% | | WebArena | 58.1% | | WebVoyager | 87% | | SimpleQA, gpt-4o-search-preview | 90% | | SimpleQA, gpt-4o-mini-search-preview | 88% | ### Key takeaways These models give access to sourced web search and interface automation from the API. They are marked as preview: reliability and safeguards (validation of sensitive actions) remain the developer's responsibility.
Links
Tags : LLM, OpenAI, Agent, Computer Use, Recherche web, API Responses