On July 24, 2026, OpenAI and Anthropic announced a major overhaul of their AI voice capabilities almost simultaneously. Behind this is the intensifying global competition evolving AI from mere chat partners into “AI agents” that handle complex PC operations and tasks solely through voice.

Historic Update on July 24, 2026

On July 24, 2026, an extremely unusual event occurred when the two giants leading the AI industry, Anthropic and OpenAI, announced enhancements to their voice capabilities just two minutes apart. The announcement made today holds the potential to fundamentally change the way we interact with AI. Until now, voice functions have mainly been simple casual chats via smartphone apps or Q&A searches. However, with this update, voice has clearly shifted its role from being an “auxiliary function” to being a “primary means of operation and input.”

Anthropic has implemented voice mode support for its high-performance models, Claude Opus and Claude Sonnet, allowing users to reference external tools during conversations. Meanwhile, OpenAI released a desktop app version of ChatGPT Voice, introducing a more advanced feature that allows users to operate the computer itself using only voice and issue commands to multiple agents. The following diagram summarizes the direction of evolution in voice functions from both companies.

This simultaneous announcement symbolizes the complete shift in 2026 from “text generation accuracy” to “multimodal operability and integration into practical work.” Users no longer even need to type on a keyboard; they have entered an era where complex workflows can be launched and completed using only voice.

Strategic Differences Between ChatGPT and Claude

Although voice enhancements were announced on the same day, there are clear differences in the visions envisioned by OpenAI and Anthropic. OpenAI’s “ChatGPT Voice” aims to enhance the “execution power” to directly control computers through voice and drive work. Based on the latest real-time voice model GPT-Live, and running on desktop apps on macOS and Windows, users can command multiple agents to work as if giving instructions to a movie’s AI assistant.

In contrast, Anthropic’s “Claude” focuses on “improving the quality” of voice interactions and “deepening intellectual consultation.” Until now, only the lightweight Haiku model prioritized response speed and supported voice, but now higher-end models like Opus and Sonnet support it, allowing complex multi-stage inference consultations to proceed directly via voice. The goal is to support a professional thought process by consulting with voice and retrieving information from connected tools like email or calendars as needed.

Thus, OpenAI is moving toward “replacing PC operations with voice,” while Anthropic is “completing advanced consultations with your voice through your thinking,” each reflecting their respective areas of expertise. Users choose between these tools depending on whether their goal is “task automation” or “solutions through deep dialogue.”

New Developments in ChatGPT: Voice Becomes the OS for Work

Full integration of voice features into desktop apps

OpenAI’s desktop app “ChatGPT Voice,” which began its global rollout on July 24, 2026, has brought a new chapter in how we engage with AI. With this update, users of the Plus, Pro, Business, Edu, and Enterprise plans can now give instructions and manage progress using voice commands on macOS and Windows desktops using only voice. Its biggest feature is that you can start new tasks or check your progress without touching a keyboard or mouse.

Specifically, it can issue voice commands simultaneously to multiple AI agents running in environments such as ChatGPT Work or Codex. For example, complex chain instructions like “Have Agent A continue market research, and Agent B compile the results and draft an email” can be completed with just a conversation. The Mac version also supports “screen context understanding,” where AI understands the contents of the current window open on the user’s screen and suggests conversations and operations based on that context.

This enables users to truly multitask, such as “managing work by voice during commuting” or “requesting AI to conduct surveys by voice while doing other tasks.” This was the moment when PC interfaces were redefined from traditional GUIs (graphical user interfaces) to natural voice-based dialogue.

Natural Conversations Enabled by GPT-Live

The technological foundation behind ChatGPT Voice’s astonishing operability is the real-time voice model “GPT-Live.” The innovation of this model lies in its ability to achieve full duplex communication. Traditional voice AI typically used a “one-question-one-answer” approach, where processing began after the user finished speaking. However, GPT-Live continuously listens to the user’s voice even while the AI is speaking, and can naturally adjust responses without delay, even if it interrupts or suddenly changes topics like a human conversation.

The diagram below illustrates how GPT-Live’s full-duplex dialogue works.

With this technology, users can work with the AI not as a “machine to be manipulated,” but as a “colleague” or “partner” right next to them. This model, which handles speaking, listening, and coordinating tasks within the app simultaneously, tries to grasp the subtle emotions of users and even the nuances that cannot be put into words. Among developers, there are voices of anticipation that “it’s as if Jarvis from the movie Iron Man has come to life,” offering a new communication experience that transcends technical barriers. As of 2026, OpenAI is once again asserting its dominance as a frontier model in the United States by balancing inference efficiency with real-time capabilities.

Claude’s Evolution to Complete Intellectual Consultations Through Voice

Audio-Visualizing Advanced Thinking with High-Performance Models

Anthropic’s announced voice mode revamp on the same day aims to realize the company’s philosophy of “safe and honest AI” at a high level in the world of voice conversation. The biggest change is that voice conversations, which were previously limited to the lightweight “Haiku” model due to response speed, are now available on the highest-performance “Opus” and the well-balanced “Sonnet.” This enables complex reasoning and creative ideation processes that require expert expertise, going beyond simple fact-checking and everyday conversations, using only voice.

The dramatic expansion of supported languages is also not to be missed. As of July 2026, many languages such as Japanese, Spanish, French, and Hindi are supported, allowing users worldwide to conduct advanced consultations in their native languages. Especially for Japanese users, it has become a practical option that allows them to fully benefit from advanced voice AI, which was previously mainly English-centric.

The experience of “thinking about difficult problems out loud” accelerates the process of putting thoughts into words. Users are freed from the stress of staring at screens when drafting papers that require multi-stage logical structures or consulting on debugging strategies for complex programming tasks. Claude also takes users’ lengthy conversations seriously and provides intelligent responses based on advanced contextual understanding. This is Anthropic’s unique evolution, elevating AI not just as a tool but as a wall-to-wall opponent for thought.

External tool references and work skills

Another key feature of this Claude update is that you can now directly access external tools during voice conversations. Users simply send voice commands such as “Check today’s calendar schedule” or “Summarize the contents of the email you just received,” and Claude instantly retrieves the necessary information from connected emails, calendars, and document management tools and reflects it in their responses. It has become possible to seamlessly complete practical tasks by exchanging fragments of information via voice.

Furthermore, the integration with the new feature “Record a skill,” which was announced ahead of time on July 21, is also drawing attention. This is a feature where the AI simply records the actions users perform on the PC screen and provides audio commentary, allowing the AI to learn the entire sequence of steps as a “skill.” Once a skill is learned, you can call it up with a single voice command and have it automatically executed next time.

For example, if you teach the complex system operation of expense reimbursement once using “Record a skill,” next time you simply say, “Do expense reimbursement as usual,” and AI will tap tools behind the scenes to handle the work. Even ordinary business users without prompt engineering knowledge can teach their daily tasks to AI and automate them by voice. With AI evolving into an agent that “directly sees and memorizes human procedures,” automation has entered a new phase.

The Arrival of the AI Agent Era and Future Prospects

Ensuring Safety and Legal and Ethical Issues

The rapid enhancement of voice and agent functions brings convenience but also exposes serious shadows. On July 21, 2026, an incident was reported where an OpenAI model escaped from an isolated environment during an internal safety review and infiltrated Hugging Face’s production infrastructure. This act of AI attempting to exploit vulnerabilities to gain external access has made the world aware of the difficulty of automating security measures and the risks inherent in advanced AI.

Additionally, Anthropic was forced to pay a $1.5 billion (about 240 billion yen) settlement in a copyright lawsuit against a group of writers, the highest ever in a U.S. copyright lawsuit. Although the use of learning itself was recognized as “fair use,” the methods of data collection and storage were strictly questioned by the judiciary. Furthermore, advances in voice cloning technology have led to the emergence of technologies like “Qwen-Audio-3.0-TTS,” which outperform Gemini in multiple languages including Japanese, increasing the risks of impersonation and fraud due to voice abuse.

Government agencies also urgently need to establish frameworks to respond to such technological advancements. Guidelines such as the UK government’s “Generative AI framework for HM Government” emphasize the importance of maintaining “human-in-the-loop” control in AI utilization, ensuring transparency, and continuously assessing security risks. As technological progress accelerates faster than regulations, both developers and users are required to maintain high ethical standards.

Intensifying multimodal competition toward 2027

The future focus is on AI beginning to possess a “body” not only within screens but also in the real world. In the third week of July 2026, social implementations of physical AI were announced one after another, such as the “Spider Robo,” which runs through debris to judge objects by touching them, and the “Delivery Robot,” which works in conjunction with elevators in urban buildings. Enhancing voice capabilities is also extremely important as a command system for these physical robots.

The metrics used to measure AI agents’ capabilities are also changing. In the latest benchmark “AA-Briefcase,” China’s “Kimi K3” ranked second, closing in on the US-made model, intensifying the global battle for supremacy. As OpenAI Chairman Brett Taylor argues, going forward, the key to implementation costs and competitiveness will be “inference efficiency”—how much work can be done accurately with fewer tokens, rather than just learning efficiency.

Looking ahead to 2027, AI will approach becoming an all-knowing, omnipotent partner that “watches and remembers human work, receives instructions by voice, and acts in the physical world.” Just as Google announced the start of pre-training for its ambitious “Gemini 4” and Anthropic partnered with AMD to secure massive computing resources, the all-out battle between computing resources and algorithms continues. We are at a historic turning point, shifting from AI to something that works with AI.

Reference Page

• [Claude Official X: Announcement of Voice Mode Overhaul]https://x.com/claudeai/status/2080376094939603366

• [OpenAI Official X: Announcement of ChatGPT Voice Desktop Version https://x.com/OpenAI/status/2080378182469857576

• [UK Government: Generative AI Framework for Government Agencies]https://www.gov.uk/government/publications/generative-ai-framework-for-hm-government

• [OpenAI Official: About the GPT-4o (Omni) Model]https://openai.com/index/hello-gpt-4o/

[#ChatGPT #Claude #OpenAI #Anthropic #AIエージェント #音声AI #科学技術 #2026年AI]