When your AI assistant hallucinates or oversimplifies a critical decision, the cost isn't measured in tokens—it's measured in revenue loss and operational risk. OpenAI o3-pro addresses this by prioritizing deep reasoning over speed, fundamentally changing how enterprises can deploy AI for high-stakes problem-solving.

OpenAI o3‑pro: The New AI Model That Thinks Longer and Performs Better

OpenAI has just unveiled OpenAI o3‑pro, an advanced version of its top-tier AI model designed to "think longer" and deliver more reliable responses. Updated in the company's latest model release notes (June 10, 2025), o3‑pro is now available to ChatGPT Pro users and via the API, bringing significant improvements in reasoning and performance. This article breaks down what o3‑pro offers and highlights a few other recent updates in OpenAI's model lineup.

---

A 'Pro' Upgrade to OpenAI's Most Advanced Model

OpenAI o3‑pro is built on the foundation of OpenAI o3, which was introduced in April 2025 as the company's most powerful reasoning model to date. Like its predecessor, o3, the new o3‑pro can use tools such as web browsing, file analysis, image understanding, Python coding, and long-term memory to augment its capabilities. This means o3‑pro doesn't just generate text; it can search the web for information, analyze uploaded documents, interpret visuals, run code, and remember context from earlier conversations to personalize responses. These abilities make it a highly versatile AI assistant for complex, multi-step tasks.

What sets o3‑pro apart is its emphasis on deep reasoning and reliability. It's tuned to "perform inference for longer and output reliable answers," prioritizing accuracy over speed. In practice, this means o3‑pro will spend more time working through a problem step-by-step, which is especially beneficial for domains like math, science, and coding where careful reasoning is required. Since its launch, users have gravitated toward the earlier Pro model (o1‑pro) for exactly these kinds of tasks, and o3‑pro continues to excel in scientific analysis, programming, and other knowledge-intensive areas. OpenAI explicitly recommends using o3‑pro for challenging questions where "waiting a few minutes is worth the tradeoff" to get a more dependable answer. In other words, if you're tackling a hard problem and can tolerate a bit more latency, o3‑pro aims to give you the best possible solution.

Outperforming Previous Models

Early evidence suggests that o3‑pro is a notable leap forward in capability. Expert evaluators consistently prefer o3‑pro's answers over the base o3 model's answers across every category tested, with especially strong wins in key fields like science, education, coding, business, and writing. Reviewers found o3‑pro's responses to be clearer, more comprehensive, more accurate, and better at following instructions than those from o3. This is a significant result, indicating that the "pro" tuning isn't just a minor tweak but yields qualitatively better output in a broad range of tasks.

Academic and benchmark evaluations echo this trend. OpenAI reports that o3‑pro outperforms both o1‑pro and o3 on rigorous benchmarks for math, science, and coding. For example, on the AIME 2024 math competition and Codeforces coding challenge, o3‑pro achieved higher scores than its predecessors, demonstrating superior problem-solving prowess. The model was even put through a strict "4/4 reliability" test - where it only passes if it can answer the same question correctly four times in a row - and o3‑pro came out on top in areas like advanced mathematics, PhD-level science questions, and competitive programming. In short, by objective measures o3‑pro appears to be the most capable and reliable ChatGPT model yet.

OpenAI o3‑pro consistently outperforms its predecessors (o1‑pro and base o3) in expert evaluations across various domains. In tests spanning science, programming, education, business, and writing, human reviewers preferred o3‑pro's answers for their greater clarity, thoroughness, and accuracy.

Not only does o3‑pro produce better answers, but it can also leverage its toolset to handle tasks that previous models might struggle with. For instance, its ability to "reason about visual inputs" means it can analyze images or charts you provide, making it useful for tasks like debugging a diagram or extracting insights from a graph - something standard text-only models cannot do. All these enhancements make o3‑pro a powerful ally for anyone building AI-driven solutions or seeking help on complex projects.

---

Availability, Pricing, and Limitations

OpenAI o3‑pro is available immediately to users on the ChatGPT Pro and Team plans, where it replaces the older o1‑pro model in the model picker. Enterprise and Education plan customers are slated to get access in the week following the release. Developers can also integrate o3‑pro via OpenAI's API as of June 10, 2025. The API usage is priced at $20 per million input tokens and $80 per million output tokens (for reference, 1 million tokens is roughly 750k words). This pricing reflects o3‑pro's position as a premium model, aimed at use cases where its advanced reasoning justifies the cost.

It's worth noting that o3‑pro uses the same underlying model architecture as the base o3 model, so OpenAI directs users to the o3 system card for full details on its safety and limitations. In practice, o3‑pro inherits o3's safety mitigations and policies, but as always with powerful AI models, users should remain vigilant for any unexpected behavior.

There are a few temporary limitations to be aware of. As of launch, OpenAI has disabled new temporary chats with o3‑pro in ChatGPT (citing a technical issue they are working to resolve). This suggests that session-based use is fine, but starting fresh, ephemeral conversations might be restricted until the fix is in place. Additionally, o3‑pro currently cannot generate images, unlike some other models - if you ask for an image in ChatGPT while using o3‑pro, it won't fulfill that request. Instead, you'd need to switch to a model like GPT-4o or OpenAI o3 (the base model) or o4-mini to use ChatGPT's image creation feature. Finally, Canvas, OpenAI's experimental AI workspace for visual brainstorming, is not supported by o3‑pro yet. These omissions are likely temporary, and future updates may expand o3‑pro's capabilities further - but for now, the focus of o3‑pro is clearly on text-based reasoning performance.

---

Other Recent Updates in OpenAI's Models

OpenAI's June 10 release of o3‑pro follows a series of rapid improvements and new model launches throughout 2025. Here are a few highlights from the model release notes leading up to o3‑pro:

• Improved GPT-4o (May 12, 2025): OpenAI updated GPT-4o's system instructions to ensure the image generation tool is invoked whenever a user requests an image in ChatGPT. This tweak helps the multimodal GPT-4o model properly hand off image creation tasks, making for a smoother experience when you ask ChatGPT to draw or visualize something.

• Fine-Tuning Fixes (April 2025): In late April, OpenAI addressed some issues with GPT-4o - on April 29 they rolled back a recent update because the model had become overly agreeable ("sycophantic") in its responses. A few days earlier (April 25), they introduced optimizations for GPT-4o to better manage its memory and improve problem-solving in STEM topics, also making it more proactive in guiding conversations. These iterative fixes show OpenAI's responsiveness in refining model behavior based on user feedback and observed quirks.

• OpenAI o3 & o4-mini Launched (April 16, 2025): OpenAI first unveiled the o3 model (the base model behind o3‑pro) in mid-April as a new state-of-the-art reasoning AI. OpenAI o3 set new benchmarks in coding, math, science, and even visual reasoning, making about 20% fewer major errors than the older OpenAI o1 model on hard real-world tasks. Alongside o3, they released OpenAI o4-mini, a smaller, fast model optimized for cost-efficient reasoning that still achieves remarkable performance for its size (even outperforming the previous o3-mini on many tasks). The introduction of o3 and o4-mini marked the beginning of this new generation of "reasoning" models focused on complex analytical tasks.

• GPT-4.5 Research Preview (Feb 27, 2025): Earlier in the year, OpenAI rolled out GPT-4.5 as a research preview to Pro users - at the time, their largest and most advanced language model for conversation. GPT-4.5 expanded the model's knowledge base and improved its ability to follow nuanced instructions, with testers noting it felt more natural and was less prone to hallucinating facts. While GPT-4.5 is still in preview, it signaled OpenAI's continuing push towards bigger and better models, setting the stage for the refined o3-series models that followed.

---

Looking Ahead

For AI founders and enthusiasts, OpenAI's o3‑pro launch is a clear sign of the rapid evolution in AI capabilities we're witnessing in 2025. Each "pro" model increment - from o1‑pro to o3‑pro - brings more depth of reasoning and reliability, opening the door for more complex and trustworthy AI-driven applications. Whether you're building an AI coding assistant, a scientific research helper, or an educational tool, the improvements in models like o3‑pro mean you can tackle harder problems with greater confidence in the AI's output.

Equally important is OpenAI's cadence of continuous improvement. The brief timeline above shows how frequently models are being tweaked, enhanced, or completely new ones are introduced. Staying on top of these release notes is becoming essential for anyone in the AI space - the latest model updates can offer new features (like better handling of images or code), fix important issues, or provide opportunities to optimize costs (as seen with efficient models like o4-mini). As OpenAI o3‑pro rolls out to more users and developers in the coming days, we can expect further feedback and refinements. In the meantime, this new model gives an exciting glimpse into the future of AI reasoning: one where long-term thinking and robust tool usage combine to solve problems that were once out of reach for chatbots.

---

Written by Dr Hernani Costa | Powered by Core Ventures

Originally published at First AI Movers.

Technology is easy. Mapping it to P&L is hard. At First AI Movers, we don't just write code; we build the 'Executive Nervous System' for EU SMEs.

Is your AI architecture creating technical debt or business equity?

👉 Get your AI Readiness Score (Free Company Assessment)