OpenAI launches o3 and o4-mini models with advanced image reasoning and tools

OpenAI launches o3 and o4-mini models with advances in visual reasoning and use of integrated tools. The company announced this Wednesday the launch of the new o series models, which represent a significant advance in processing capabilities. The o3 stands out as the most powerful in the range to date, while the o4-mini offers optimized performance at lower cost and greater speed. Esses models were trained to think longer before responding, incorporating chain reasoning with support for multiple modalities.

Models allow direct integration of images into the reasoning process. Usuários may upload low-quality diagrams, sketches, or whiteboards for analysis and manipulation. Essa functionality expands applications into technical and creative areas.

Key capabilities of the new models

OpenAI o3 leads in performance across coding, math, science, and visual perception benchmarks. Ele outperforms previous versions in tasks that require extended thinking and the use of native tools. The model combines web searching, code execution in Python, file analysis and image generation in single streams.

The o4-mini is designed for efficiency. Ele maintains a high level of accuracy in similar tasks, but with reduced latency and lower costs. Essa version serves users who need quick responses without compromising significant quality.

Both models support full tools. Incluem web navigation, file analysis, automations and contextual memory for more consistent interactions.

Availability and initial access

ChatGPT Plus, Pro and Team gained immediate access to the models in the options selector. The o3 appears as the main choice for complex tasks, while the o4-mini and high-performance variants replace previous options. The API release occurs gradually for developers.

Free users may experience limited capabilities soon. The company prioritizes controlled rollout to ensure stability and collect feedback.

Advances in reasoning with images

OpenAI highlights the ability to “think with images” as a differentiator. Models not only describe visual content, but integrate image information directly into problem-solving logic. Isso allows you to manipulate, crop or transform visual elements during processing.

Examples include analysis of technical diagrams or manual sketches. The system identifies spatial relationships and applies step-by-step reasoning to reach accurate conclusions.

This innovation opens up avenues for applications in engineering, education and scientific research. Profissionais may submit flowcharts or notes for detailed explanations or corrections.

Leia Tambem

Colby Minifie confirms Ashley Barrett’s powers in The Boys season five

Napoli x Milan in Serie A with confirmed lineups and where to watch live

New battery test puts Galaxy S26 Ultra ahead of iPhone 17 Pro Max in global ranking

Performance in benchmarks and comparisons

o3 sets new records in independent coding and advanced mathematics assessments. Ele demonstrates superiority in problems that require multiple logical steps and internal verification. Resultados shows significant gains over its predecessor o1 in standardized metrics.

The o4-mini balances performance and efficiency. Ele achieves scores close to o3 in selected tasks, but with much lower resource consumption. Essa optimization makes it easier to use at scale for businesses and individual developers.

Security assessments indicate that both models maintain resilience against attempts to bypass restrictions. Eles consistently refuse harmful content.

Integration with tools and ecosystem

Models natively incorporate tools into reasoning. Isso includes Python code execution for complex calculations and web searching for up-to-date data. The combination allows you to solve real problems that require multiple sources and verifications.

Tools like file analysis and image generation expand usefulness. Usuários create complete flows in a single interaction, from searching to viewing results.

The company has released complementary tools for programmers. Elas facilitate integration into development environments and accelerate workflows.

Security measures and assessments

OpenAI applied updated preparedness framework to assess risks. The models did not meet high thresholds in critical categories such as biological capabilities, cybersecurity, or self-improvement. Independent review confirmed proper alignment.

Measures include rigorous testing to refuse harmful content and resistance to jailbreaks. The company continues to monitor production use for necessary adjustments.