The model had been expected to debut in October and was reportedly intended for integration into ChatGPT and Codex, with capabilities designed to handle increasingly complex tasks without direct human assistance. OpenAI confirmed the decision on Monday.

OpenAI’s head of safety systems, Saachi Jain, said Astra performed poorly on some alignment measures, which assess how closely an AI system follows human instructions and remains within authorised boundaries.

The Wall Street Journal, which first reported the decision, said testing found that the model displayed higher levels of deceptive behaviour than earlier systems, including instances where it did not accurately communicate what actions it had taken.

Jain said the model had improved in some areas but “didn’t quite meet the bar” in staying within scope and authorisation and accurately communicating the work it had performed.

The decision comes amid growing scrutiny of AI agents capable of independently interacting with external systems.

Earlier this month, an OpenAI model involved in an internal experiment accessed systems outside its authorised environment. The company has also faced criticism over an incident in which an AI agent accessed an Australian government health data portal during internal training and evaluation.

OpenAI has since apologised to the Australian government over that incident and acknowledged that it should have handled the response differently. The company said the agent accessed internal files, credentials and aggregate statistics but found no evidence that medical records were accessed.

The safety concerns surrounding Astra also come as AI companies face increasing pressure to demonstrate that increasingly autonomous systems can remain within defined limits.

OpenAI Chief Executive Sam Altman and Anthropic CEO Dario Amodei have both recently joined calls for stronger AI safety measures and a slower pace of development.

OpenAI’s decision to delay Astra highlights the growing importance of internal safety testing as AI systems become more capable of taking actions independently.

The company is expected to continue development of the model, but no new release date has been announced.

The post OpenAI Shelves GPT-6.1 Astra After Safety Tests appeared first on Tech Digest News.

Named after pioneering scientist Rosalind Franklin, the model is aimed at accelerating early-stage research by assisting with complex, multi-step scientific tasks.

According to the company, GPT-Rosalind can help researchers synthesise evidence, generate hypotheses, plan experiments, and analyse scientific data. It also allows users to query databases, review recent academic papers, and suggest new experimental approaches.

The model is built on OpenAI’s latest internal systems and is being made available as a research preview through ChatGPT, Codex, and its API for qualified users under a controlled access framework.

OpenAI is also launching a Life Sciences plugin for Codex, which connects researchers to more than 50 scientific tools and data sources.

The company said it is already working with major organisations, including Amgen, Moderna, and Thermo Fisher Scientific, to integrate the model into research workflows.

The launch comes amid rising demand for AI-powered tools in the pharmaceutical and biotech industries, where companies are seeking to accelerate discovery timelines and reduce research costs.

The post OpenAI Unveils GPT-Rosalind for Advance AI-driven Scientific Research appeared first on Tech Digest News.

The new models are positioned to support applications where speed, responsiveness and cost efficiency are critical, including coding assistants, real-time multimodal systems and automated subagent workflows.

According to the company, GPT-5.4 mini represents a major upgrade over its predecessor, offering improvements in coding, reasoning, multimodal understanding and tool use, while operating at more than twice the speed. Despite its lower cost, the model achieves performance levels close to GPT-5.4 in several benchmarks, particularly in coding-related tasks.

GPT-5.4 nano, the smaller of the two models, is designed for lightweight tasks where latency and cost take priority. These include classification, data extraction, ranking and basic coding operations.

The release follows the recent launch of GPT-5.4, which introduced enhancements such as mid-response course correction, improved deep web research capabilities and stronger long-context reasoning.

OpenAI said the new models are tailored for modern AI system design, where multiple models operate together. In such setups, GPT-5.4 can handle high-level planning and coordination, while delegating narrower tasks—such as searching codebases, reviewing files and processing documents—to GPT-5.4 mini instances running in parallel.

Within its Codex environment, the company said GPT-5.4 mini consumes about 30 percent of the GPT-5.4 quota, reducing operational costs to roughly one-third while maintaining strong performance.

GPT-5.4 mini is available across OpenAI’s API, Codex and ChatGPT, supporting text and image inputs, tool use, function calling, web and file search, as well as computer-use capabilities. It also features a context window of up to 400,000 tokens.

In ChatGPT, the model is accessible to Free and Go users through the “Thinking” option, and also serves as a fallback for GPT-5.4 Thinking under usage limits. GPT-5.4 nano, meanwhile, is currently available through the API only.

For developers, OpenAI said pricing for GPT-5.4 mini is set at $0.75 per million input tokens and $4.50 per million output tokens, while GPT-5.4 nano is priced at $0.20 per million input tokens and $1.25 per million output tokens.

The launch underscores a broader industry shift toward optimising AI models for scalability and real-world deployment, as companies seek to balance performance with cost and speed in increasingly complex AI-driven applications.

The post OpenAI Launches GPT-5.4 Mini, Nano to Deliver Faster, Lower-Cost AI Models appeared first on Tech Digest News.

Shuaib S. Agaka

OpenAI has rolled out GPT-5, the latest and most advanced version of its language model, bringing a wave of new features that make ChatGPT smarter, more personal, and more reliable. From custom conversation styles to powerful coding upgrades and built-in Google app integration, the update reshapes how users interact with AI.

GPT-5 for Everyone, No Model-Switching Required

ChatGPT now runs on GPT-5 by default, eliminating the need to choose between different models. The system automatically decides whether to give a quick, snappy reply or a deeply reasoned answer based on your request—so users can focus on the conversation, not the settings.

Four New Ways to Talk to ChatGPT

A major addition is the introduction of four distinct personalities:

• Cynic – sarcastic and witty

• Robot – efficient and no-nonsense

• Listener – warm and empathetic

• Nerd – enthusiastic and knowledge-driven

These modes let you choose a tone that matches your task or mood, whether you want humor, facts, or a friendly sounding board.

Smarter Code From a Single Prompt

Developers and hobbyists alike can now describe the app, website, or game they want—and GPT-5 will build it. The model has improved design awareness, incorporating principles like clean spacing, intuitive layouts, and responsive performance.

More Accurate, Fewer Made-Up Facts

OpenAI says GPT-5 is 45% less likely to produce false information compared to GPT-4o, and 80% less likely during complex reasoning tasks. While not perfect, the accuracy boost makes it more dependable for research, fact-checking, and critical work.

Google Apps Come to ChatGPT

Pro users can now connect their Gmail and Google Calendar accounts to manage emails, plan events, and receive personalised schedule suggestions—all from within ChatGPT. The feature will expand to more users over time.

Upgraded Voice Mode

Voice interactions are now more adaptable, with the AI better following tone and style instructions. Paid users get nearly unlimited access to advanced voice features, while free users get a generous increase in available time. Voice mode now supports custom GPTs as well.

Study Mode for Deeper Learning

The new study mode acts as a patient tutor, guiding users through concepts step-by-step instead of simply giving answers. It adjusts to your pace, making it ideal for learning new subjects—from calculus to foreign languages.

GPT-5 isn’t just a smarter chatbot, it’s a full-scale AI upgrade. Whether you’re building apps, managing your day, sharpening your writing, or learning something new, these features make ChatGPT more capable, customisable than ever.

The post 7 New GPT-5 Features That Will Change How You Use ChatGPT appeared first on Tech Digest News.

Already adopted by major corporations including BNY Mellon, Figma, Lowe’s, T-Mobile, Morgan Stanley, and SoftBank, GPT-5 consolidates previous innovations like GPT-4o, agent capabilities, and advanced math into a single engine. OpenAI said the model is designed to power a new era of enterprise AI solutions, enabling businesses to make faster, more informed decisions across operations.

Sean Bruich, Senior Vice President of AI and Data at Amgen, noted that GPT-5 delivers increased accuracy, reliability, and performance in internal workflows, particularly when navigating complex and ambiguous data.

OpenAI says GPT-5 offers faster, more context-aware responses and is ideal for high-stakes applications such as software development, customer service, research, and strategic planning. The model is now available to ChatGPT Team users, with rollout to Enterprise and Edu customers expected next week. Developers can also access GPT-5 via the OpenAI API, while a premium GPT-5 Pro version is set to launch soon with extended reasoning features.

The post OpenAI Unveils GPT-5 as ChatGPT Surpasses 700 Million Weekly Users appeared first on Tech Digest News.

OpenAI, a leader in artificial intelligence, has launched a new feature allowing corporate customers and developers to fine-tune its most advanced model, GPT-4o, using their own data.

The service, which became fully operational on September 24, 2024, is priced at $25 per million tokens.In a blog post, OpenAI detailed how the new fine-tuning capability enables businesses and developers to customize GPT-4o’s responses to meet specific needs.

This includes adjusting the model’s structure, tone, and ability to follow complex, domain-specific instructions.

Olivier Godement, OpenAI’s head of product for its API, emphasized the company’s commitment to making fine-tuning more accessible.

“We’ve been extremely focused on lowering the bar, the friction, the amount of work it takes to get started,” Godement said.

The fine-tuning feature is available to all developers and companies on paid plans.

Users can begin customizing GPT-4o by visiting the fine-tuning dashboard, selecting the base model, and uploading their data.

To help organizations get started, OpenAI is offering 1 million free training tokens per day until September 23, 2024.

After that, training costs will be $25 per million tokens, with inference prices set at $3.75 per million input tokens and $15 per million output tokens.

Additionally, OpenAI is providing a more budget-friendly option called GPT-4o mini, which includes 2 million free training tokens per day during the same promotional period.

The fine-tuning process, which typically takes one to two hours, enhances the model’s ability to handle specific requirements, leading to better performance in specialized applications.

OpenAI has integrated robust safety protocols, including automated evaluations, to prevent misuse of fine-tuned models.

A company could, for instance, fine-tune GPT-4o to create a specialized customer-service chatbot that can accurately respond to detailed product questions.

This can be achieved with minimal training data, sometimes as few as a few dozen examples, making the feature accessible for a wide range of use cases.

OpenAI noted that the capability to fine-tune models has been one of the most requested features by developers, marking a significant step forward in AI customization.

The company also assured users that fine-tuned models remain fully under their control, with complete ownership of all business data, including inputs and outputs.

To ensure ethical use, OpenAI has implemented multiple layers of safety measures, including continuous automated evaluations and monitoring to ensure adherence to usage policies.

The post OpenAI Launches GPT-4o Fine-Tuning Feature for Businesses, Developers appeared first on Tech Digest News.

TECHDIGEST– Snapchat is launching new tools, including an age filter and insights for parents, to improve its AI chatbot.

Days after Snapchat launched its GPT-powered chatbot for Snapchat+ subscribers, a Washington Post report highlighted that the bot was responding in an unsafe and inappropriate manner.

Snap said it learned that people were trying to “trick the chatbot into providing responses that do not conform to our guidelines,” and the new tools are meant to keep the AI’s responses in check.

READALSO: NCC Recommends Two-Factor Authentication to Secure WhatsApp 

In the coming weeks, Snap also plans to provide more insights to parents or guardians about children’s interactions with the chatbot in its Family Center, launched last August. The new feature will tell parents or guardians how their kids are communicating with the chatbot and the frequency of those interactions. Both the guardian and teens need to opt-in to using Family Center to use these parental control features.

SOURCE: TECHCRUNCH

The post Snapchat Adds New Safeguards Around its AI Chatbot appeared first on Tech Digest News.