o3 Pro

OpenAI announced o3-pro on June 10, 2025, as a higher-compute version of its o3 reasoning model designed to think longer and deliver the most reliable responses. Like o1-pro before it, o3-pro is positioned as the most dependable option in the o-series, and is available to Pro users in ChatGPT and through the OpenAI API. The page also notes that the underlying o3 and o4-mini models are now deprecated in the API. The o3 and o4-mini release post establishes the o-series as reasoning models trained to think before responding, with first-time agentic tool use across web search, Python analysis, visual reasoning, and image generation. OpenAI states o3 set new state-of-the-art results on Codeforces, SWE-bench, and MMMU benchmarks, framing o3-pro as the reliability-focused variant built on that stronger reasoning foundation. The update paragraph is the sole direct creator evidence about o3-pro on the page.

OpenAI launched o3-pro on June 10, 2025, positioning it as its most capable model and a higher-compute version of the o3 reasoning model that works through problems step by step. The model became available to ChatGPT Pro and Team users that day, replacing o1-pro, with Enterprise and Edu access following the next week. It also went live in OpenAI's developer API the same afternoon, marking the first time Pro-tier reasoning reached developers through the API. According to the TechCrunch report, o3-pro is priced at $20 per million input tokens and $80 per million output tokens, a significant premium reflecting its reasoning focus. The model supports the same tool set that makes ChatGPT useful, including web search, file analysis, visual reasoning, Python execution, and personalization via memory. OpenAI noted that full safety details are covered in the o3 system card since o3-pro shares its underlying model.

OpenAI released o3-pro on June 10, 2025 as part of the o3 model family, building on the earlier o1-pro release and targeting complex reasoning in physics, mathematics, and software development. According to a SPLX head-to-head evaluation, the model uses reinforcement-learning reasoning to work through problems step by step for logic-intensive enterprise tasks. The evaluation deployed o3-pro and GPT-4o as assistants helping customers choose insurance policies under benign and adversarial conditions. In that SPLX red-team test, GPT-4o outperformed o3-pro on speed and reliability while also proving more cost-efficient, with lower failure rates and a smoother user experience. The findings challenge OpenAI's positioning of o3-pro as its most advanced commercial offering, suggesting its heavier reasoning overhead can hurt practical enterprise deployments. Results are limited to the specific insurance-assistant workload and do not generalize across all reasoning domains.

OpenAI's official API documentation confirms o3-pro is a higher-compute version of o3 designed to think harder and deliver consistently better answers on complex reasoning tasks. The model is available only through the Responses API to support multi-turn interactions before responding, and is currently marked Deprecated. It uses a 200,000-token context window with up to 100,000 max output tokens, and requests may take several minutes, so background mode is recommended to avoid timeouts. The docs list pricing at $20 per million input tokens and $80 per million output tokens, with text and image input support but audio unsupported. Knowledge cutoff is June 1, 2024. The endpoint listing shows Responses v1/responses and Chat Completions v1/chat/completions support, while live v1/live/sessions is not listed for o3-pro. Quick-comparison pricing shows o3-pro at roughly 10x the input cost of base o3 ($2/M) and o3-mini ($1.10/M), reflecting the additional inference compute invested per response.