OpenAI Releases Three New Voice Models for AI Agents and Translation

OpenAI has released three new realtime audio models for developers, including GPT-Realtime-2 for conversational AI agents, a live translation system supporting more than 70 languages, and a streaming speech-to-text model.

By Daniel Mercer Edited by Maria Konash Published:
OpenAI launches GPT-Realtime-2, live translation, and streaming speech-to-text models for AI voice apps. Image: OpenAI

OpenAI has introduced three new realtime audio models through its API platform, expanding its push into conversational AI agents and voice-based software interfaces. The release includes GPT-Realtime-2, a new voice model with GPT-5-level reasoning capabilities, GPT-Realtime-Translate for live multilingual speech translation, and GPT-Realtime-Whisper for low-latency streaming transcription.

The company said the models are designed to support a new generation of voice applications capable of reasoning through requests, using external tools during conversations, translating speech live, and handling continuous spoken interaction in real time.

GPT-Realtime-2 is positioned as OpenAI’s most advanced voice interaction model so far. The system supports live conversational workflows where the AI can process interruptions, maintain long context windows, call tools in parallel, and continue conversations naturally while tasks are being completed in the background.

OpenAI expanded the model’s context window from 32,000 to 128,000 tokens and introduced adjustable reasoning levels ranging from minimal to “xhigh,” allowing developers to balance latency against reasoning depth. The company said GPT-Realtime-2 scored 96.6% on the Big Bench Audio Intelligence benchmark, compared with 81.4% for GPT-Realtime-1.5.

The company also introduced GPT-Realtime-Translate, a live speech translation model supporting more than 70 input languages and 13 output languages. OpenAI said the model is designed for customer support, international business communication, events, education, and multilingual voice interfaces where conversations need to continue naturally across languages without noticeable delays.

GPT-Realtime-Whisper, meanwhile, focuses on streaming speech recognition. The model transcribes spoken audio as conversations happen, allowing developers to build live captioning systems, meeting assistants, support tools, and voice-driven enterprise workflows with lower latency.

OpenAI said companies including Zillow, Intercom, Priceline, Deutsche Telekom, and Vimeo have already tested the new models in production-oriented voice systems.

“What stood out about GPT-Realtime-2 was the intelligence and tool-calling reliability it brings to complex voice interactions,” said Zillow SVP and Head of AI Josh Weisberg, who said the model improved call success rates during adversarial testing.

OpenAI Pushes Beyond Text-Based Interfaces

The release reflects OpenAI’s broader strategy of moving AI interaction away from chat windows and toward continuous voice-based systems integrated directly into software products and workflows.

Rather than functioning as simple speech interfaces layered on top of chatbots, the new models are designed to operate as realtime agents capable of reasoning, retrieving information, executing actions, and maintaining conversational continuity simultaneously.

OpenAI described three emerging categories for voice AI systems:

  • voice-to-action workflows where agents complete tasks directly from spoken instructions,
  • systems-to-voice interfaces where software proactively communicates updates through speech,
  • voice-to-voice interactions involving live multilingual translation between users.

The company highlighted examples such as AI travel assistants capable of managing itinerary changes conversationally and multilingual customer service systems that translate discussions in real time while preserving natural speech flow.

OpenAI also emphasized production safeguards around the Realtime API, including active classifiers that can interrupt sessions violating safety policies and support for additional developer-defined guardrails through the Agents SDK.

The models are available immediately through OpenAI’s Realtime API. GPT-Realtime-2 is priced at $32 per million audio input tokens and $64 per million output tokens, while GPT-Realtime-Translate costs $0.034 per minute and GPT-Realtime-Whisper costs $0.017 per minute.

AI & Machine Learning, News

OpenAI Launches $4 Billion Enterprise AI Deployment Venture

OpenAI is creating a new enterprise deployment company backed by more than $4 billion in initial funding and acquiring AI consulting firm Tomoro.

By Maria Konash Published:
OpenAI forms a $4B AI deployment firm and acquires Tomoro to expand enterprise implementation services. Image: Levart_Photographer / Unsplash

OpenAI is creating a new enterprise-focused business called OpenAI Deployment Company with more than $4 billion in initial committed investment. The company also announced the acquisition of Tomoro, a consulting firm specializing in enterprise AI implementation, as it accelerates efforts to expand adoption of ChatGPT and other OpenAI systems inside large organizations.

According to OpenAI, the new unit will help companies build and deploy AI systems by embedding specialized engineers and deployment teams directly within customer organizations. These teams will work alongside corporate departments to identify operational areas where AI systems can automate workflows, improve productivity, or support decision-making.

The acquisition of Tomoro will immediately add around 150 AI engineers and deployment specialists to the business. Tomoro was established in 2023 through a partnership aligned with OpenAI and has worked with companies including Mattel, Red Bull, Tesco, and Virgin Atlantic.

OpenAI said the deployment venture is structured as a multi-year partnership between OpenAI and 19 investment firms. The initiative is led by TPG, with Advent International, Bain Capital, and Brookfield Asset Management acting as co-lead founding partners.

The launch comes as OpenAI intensifies its enterprise expansion efforts following widespread consumer adoption of ChatGPT. The company has increasingly focused on securing long-term corporate contracts and integrating AI systems into business operations at scale.

OpenAI Moves Beyond Software Licensing

The creation of OpenAI Deployment Company signals a broader shift in how frontier AI firms are approaching enterprise adoption. Instead of only selling access to AI models through APIs or subscriptions, OpenAI is building a dedicated implementation business designed to help customers operationalize AI across complex organizations.

The strategy reflects a growing reality in enterprise AI: deploying advanced models often requires extensive customization, workflow integration, governance planning, and technical support. Many companies lack internal expertise to manage those processes independently.

By embedding engineers directly inside client organizations, OpenAI is adopting an approach closer to enterprise consulting and systems integration firms than conventional software vendors. The model resembles how companies such as Palantir Technologies work with customers to integrate AI and data systems into operational workflows.

Enterprise AI Services Become A Competitive Battleground

The announcement also highlights increasing competition between leading AI companies for enterprise market share. OpenAI’s expansion comes as Anthropic continues gaining traction with its Claude models across corporate customers.

As previously reported, both OpenAI and Anthropic were exploring acquisitions of AI deployment and consulting firms as part of broader enterprise strategies. The sector has become strategically important because large organizations often require ongoing implementation support rather than standalone model access.

The new venture also gives OpenAI a larger operational footprint inside customer environments, potentially strengthening long-term relationships and increasing dependence on its infrastructure and models.

SoftBank Explores $100B AI Data Center Investment in France

SoftBank is reportedly evaluating a multibillion-dollar AI data center project in France following talks with President Emmanuel Macron.

By Olivia Grant Edited by Maria Konash Published:
SoftBank weighs major AI data center investment in France after talks between Masayoshi Son and Macron. Image: Anthony Choren / Unsplash

SoftBank CEO Masayoshi Son has reportedly held discussions with French President Emmanuel Macron regarding a large-scale AI data center initiative in France.

According to the Bloomberg report, the project could involve a multibillion-dollar investment aimed at expanding France’s artificial intelligence infrastructure capabilities. While earlier discussions reportedly referenced potential investments of up to $100 billion, sources indicated the final amount may be lower depending on other capital commitments.

The proposal was initially raised by Macron during a meeting with Son in Tokyo, highlighting France’s efforts to attract major AI infrastructure investments and strengthen Europe’s position in the global AI race. An official announcement could reportedly come within weeks.

The discussions reflect intensifying competition among governments to secure AI infrastructure projects, particularly large-scale data centers that support model training and cloud computing. Europe has increasingly emphasized technological sovereignty and domestic AI capacity as reliance on foreign cloud providers grows.

For SoftBank, the potential investment would further expand its role in global AI infrastructure following a series of large-scale bets on semiconductors, data centers, and AI companies. The company has been actively positioning itself to benefit from rising demand for computing power and next-generation AI systems.

The project also signals how AI infrastructure is becoming a strategic geopolitical priority, with governments directly engaging technology investors to accelerate domestic capabilities and attract long-term capital into the sector.

 The discussions come as SoftBank reportedly prepares a new AI and robotics venture focused on automating data center construction, targeting growing global demand for AI infrastructure and a potential $100 billion IPO.

AI & Machine Learning, Cloud & Infrastructure, News, Startups & Investment

EU Holds AI Access Talks With OpenAI and Anthropic

The European Commission is in ongoing talks with OpenAI and Anthropic over AI model access, transparency, and regulatory engagement.

By Samantha Reed Edited by Maria Konash Published: Updated:
EU in talks with OpenAI and Anthropic over AI model access and regulatory cooperation. Image: Christian Lue / Unsplash

OpenAI said it will provide the European Union with access to GPT-5.5-Cyber, a cybersecurity-focused version of its latest AI model, as Brussels expands oversight of advanced AI systems and cyber capabilities. The announcement marks a significant step in the EU’s effort to directly evaluate frontier AI models before wider deployment.

According to OpenAI, European businesses, governments, cybersecurity authorities, and institutions including the EU AI Office will receive access to the model as part of a limited preview program for vetted cybersecurity teams. The company said the initiative falls under a broader “OpenAI EU Cyber Action Plan” focused on improving defensive cyber capabilities across Europe.

The move comes roughly one month after Anthropic introduced Mythos, its advanced cybersecurity-focused model that sparked concerns among governments and enterprises because of its ability to identify software vulnerabilities at large scale. Unlike OpenAI, Anthropic has not yet granted EU institutions preview access to the model.

European Commission spokesperson Thomas Regnier confirmed that OpenAI has already exchanged information with the Commission and that additional meetings are scheduled this week to discuss access and oversight arrangements.

“We welcome OpenAI’s transparency and intent to give commission access to new model,” Regnier said during a press briefing. He added that the arrangement would allow European authorities “to follow deployment of the model very closely, and address security concerns.”

Regnier said the Commission has also held “four or five” meetings with Anthropic regarding Mythos, but discussions remain at an earlier stage.

“The discussions with Anthropic are not yet at the same stage as the solution we have on the table from OpenAI,” he said.

OpenAI framed the agreement as part of a broader push toward collaborative oversight of cyber-capable AI systems.

“AI labs like ours shouldn’t be the sole arbiters of cyber safety as resilience depends on trusted partners working together,” said George Osborne, OpenAI’s head of OpenAI for Countries.

He added that advanced cyber models “should be available for Europe’s many defenders, not just the few.”

Europe Pushes For Direct Access To Frontier AI Systems

The agreement reflects how European regulators are moving beyond traditional compliance discussions and toward direct technical engagement with frontier AI models. Rather than relying only on company disclosures, EU institutions are increasingly seeking hands-on access to evaluate deployment risks, security implications, and potential misuse scenarios.

Cybersecurity models have become a particular focus because of their dual-use nature. Systems capable of identifying vulnerabilities can help defenders secure critical infrastructure, but they can also accelerate offensive cyber operations if misused.

OpenAI’s decision to provide access gives the company a more cooperative position with European regulators at a time when AI oversight under the EU AI Act is becoming more aggressive and technically detailed.

Anthropic Faces Growing Pressure Over Mythos Access

Anthropic’s reluctance to provide similar access to Mythos could increase regulatory pressure as governments demand greater visibility into frontier AI systems with national security implications.

Mythos has already attracted attention because of reports that it uncovered tens of thousands of software vulnerabilities, including flaws in widely used infrastructure and legacy systems. Anthropic has restricted access to the model because of concerns that the findings could be exploited by malicious actors.

The divergence between OpenAI and Anthropic also highlights an emerging split in how leading AI labs approach government oversight. Some companies are moving toward structured collaboration with regulators and security agencies, while others remain more cautious about sharing highly capable cyber models before safeguards and patching processes are in place.

Tether Data Launches Medical AI Models Designed for Smartphones and Laptops

Tether Data has introduced QVAC MedPsy, a family of compact medical AI models designed for smartphones, laptops, and edge devices. The company claims the models outperform significantly larger healthcare-focused systems while reducing inference costs and token usage.

By Laura Bennett Edited by Maria Konash Published:
Tether Data launches QVAC MedPsy AI models for edge devices with strong clinical performance and lower costs. Image: Tether

Tether’s AI research division, Tether Data, has released QVAC MedPsy, a new family of text-only medical language models optimized for edge deployment. The models come in 1.7 billion and 4 billion parameter versions and are designed to run on consumer hardware including smartphones, laptops, and wearable devices while maintaining strong medical reasoning performance.

According to Tether Data, the smaller QVAC MedPsy-1.7B model achieved an average score of 62.62 across seven closed-ended medical benchmarks. The company said this outperformed Google’s MedGemma-1.5-4B-it model by more than 11 points despite using less than half the parameters. It also approached the performance of larger reasoning-focused models such as Qwen3-4B-Thinking-2507.

The larger QVAC MedPsy-4B model reportedly surpassed MedGemma-27B-text-it on several benchmarks tied to practical healthcare reasoning. On HealthBench Hard, which measures performance in more realistic clinical scenarios, Tether Data reported scores of 58.00 for MedPsy-4B compared with 42.00 for Google’s 27-billion-parameter system.

Tether Data also emphasized inference efficiency as a major differentiator. The company said MedPsy-4B generated benchmark answers using an average of roughly 909 tokens, compared with approximately 2,953 tokens for Qwen3-4B-Thinking-2507. Lower token usage reduces latency and compute costs, which is particularly important for real-time deployment on lower-power devices.

The models are being released under the Apache 2.0 license for research and educational use. Tether Data is also publishing GGUF versions compatible with llama.cpp and its own QVAC SDK, including quantized variants designed to reduce storage requirements while maintaining most benchmark performance. The company said some compressed versions cut file size by nearly 70% with minimal performance degradation.

QVAC MedPsy was evaluated across eight benchmark suites covering clinical reasoning, biomedical research, health literacy, and underserved healthcare contexts. These included MedQA-USMLE, MedMCQA, PubMedQA, AfriMedQA, and HealthBench.

Smaller Medical Models Target Real-World Deployment

The release reflects growing demand for medical AI systems that can run locally instead of relying entirely on cloud infrastructure. Most high-performing healthcare language models are too large to deploy directly on edge devices, limiting their use in low-connectivity or privacy-sensitive environments.

By reducing parameter count and token usage while maintaining benchmark performance, Tether Data is targeting practical deployment scenarios such as offline clinical assistance, medical education tools, and decision-support systems operating directly on consumer hardware.

The focus on local inference is also significant for healthcare providers dealing with strict data privacy requirements. Running models directly on devices can reduce the need to transmit sensitive patient information to remote servers, which may simplify compliance and improve response times.

Tether Expands Into AI-Driven Health Technologies

The MedPsy launch is part of a broader expansion of Tether’s AI and health technology efforts. Earlier, the company introduced BrainWhisperer, a brain-computer interface system designed to convert neural activity into text using on-device AI processing. Tether claimed the system achieved up to 98.3% accuracy while keeping neural data local to the device.

Tether has also been expanding into consumer wellness technologies through investment activity. Eight Sleep recently received a strategic investment from Tether Investments at a reported $1.5 billion valuation, with a focus on AI-driven sleep monitoring and personalized health intelligence.

AI & Machine Learning, News

Elon Musk Merges xAI Into SpaceX Under New SpaceXAI Structure

Elon Musk said xAI will cease operating as an independent company and become fully integrated into SpaceX under a new SpaceXAI structure. The move combines Musk’s AI models, compute infrastructure, and aerospace operations into a single organization.

By Maria Konash Published: Updated:
Elon Musk merges xAI into SpaceX under SpaceXAI, combining Grok, AI supercomputers, and orbital compute plans into one organization. Image: SpaceX

Elon Musk said xAI will no longer operate as an independent business and will instead be fully integrated into SpaceX under a new structure called SpaceXAI. The consolidation combines Musk’s AI models, social platform infrastructure, supercomputing operations, and aerospace systems into a single organization.

The announcement came alongside a new compute agreement between SpaceXAI and Anthropic. Under the deal, Anthropic will gain access to Colossus 1, a large AI supercomputer cluster originally developed by xAI. Musk confirmed on X that “xAI will be dissolved as a separate company” and that products including Grok will continue under the SpaceXAI name.

The restructuring follows an earlier all-stock transaction in which SpaceX acquired xAI at valuations reportedly placing SpaceX near $1 trillion and xAI around $250 billion. The combined structure is valued at roughly $1.25 trillion.

Under SpaceXAI, Grok development, AI infrastructure, and future compute projects will operate directly within SpaceX management. The move also places major infrastructure systems such as the Colossus supercomputer clusters under the same organization responsible for launch systems, satellite operations, and Starlink.

Musk acknowledged operational issues inside xAI before the restructuring, saying publicly that the company “was not built right first time around.” The reorganization follows months of executive departures at xAI, where nearly all original co-founders had reportedly left by March.

Anthropic’s agreement with SpaceXAI includes access to more than 300 megawatts of compute capacity through Colossus 1, including over 220,000 NVIDIA GPUs spanning H100, H200, and GB200 accelerators. Musk said discussions with Anthropic leadership convinced him the company was approaching AI development responsibly, though he added that SpaceXAI reserves the right to reclaim compute capacity if systems “engage in actions that harm humanity.”

SpaceXAI Centralizes Compute And Infrastructure

The consolidation gives Musk direct control over a vertically integrated AI stack that includes compute infrastructure, model development, satellite networking, and launch systems. Analysts said combining those operations under one structure could simplify capital deployment and accelerate expansion of AI infrastructure projects.

The integration is particularly significant because AI companies are increasingly constrained by access to power, GPUs, cooling systems, and data center capacity. SpaceX already controls launch infrastructure, satellite communications, and large-scale engineering operations, which could become strategically valuable if AI compute continues expanding at current rates.

The move also changes the role of Colossus infrastructure inside Musk’s AI strategy. Leasing large portions of Colossus 1 to Anthropic allows SpaceXAI to monetize existing GPU assets while focusing internal development on newer systems such as the planned Colossus 2 cluster.

Orbital Compute Becomes A Core Strategy

SpaceXAI is also pushing more aggressively into orbital AI infrastructure. Earlier plans outlined the possibility of space-based data centers powered by solar energy and supported through Starship launches, with the goal of overcoming terrestrial limitations tied to electricity availability, cooling requirements, and land use.

The restructuring places those orbital compute ambitions directly under the same organization operating Starlink and reusable launch systems. That integration could allow SpaceXAI to coordinate launch cadence, satellite networking, power systems, and AI infrastructure development more tightly than conventional cloud providers.

Exit mobile version