OpenAI Launches ChatGPT Work as AI Agent Rivalry Escalates

OpenAI has launched ChatGPT Work, a new AI "super app" that integrates its chatbot with the Codex coding tool, enabling non-technical users to autonomously generate documents, presentations, and websites. Unveiled on July 9, 2026, the product is powered by GPT-5.6, a model family released the same day after a US government security review delayed its original June rollout. The move marks a significant escalation in OpenAI’s rivalry with Anthropic, which has rapidly expanded its own autonomous agent, Claude Cowork, in recent months.

For enterprises worldwide, the implications are substantial. ChatGPT Work and Claude Cowork represent the next evolution in workplace AI—tools designed to execute multi-step, white-collar tasks without human oversight. With both OpenAI and Anthropic reportedly preparing for public listings, the competition to dominate this market could reshape productivity software, redefine corporate AI spending, and push regulators to address the security risks posed by increasingly advanced models.

A delayed debut amid security concerns

GPT-5.6’s release was originally scheduled for June 2026, but OpenAI paused the launch after the US government requested a review, citing concerns that the model’s code-exploitation capabilities could be weaponized. The delay reflects a broader trend of heightened scrutiny over frontier AI systems, particularly those with potential dual-use applications. Sam Altman, OpenAI’s CEO, confirmed that Sol, the flagship tier of GPT-5.6, is 54% more token-efficient than competing models, a metric that could provide OpenAI with a competitive edge in cost-conscious enterprise environments.

The GPT-5.6 family is structured into three tiers: Sol (high-end), Terra (mid-range), and Luna (low-cost). Max Weinbach, an analyst at Creative Strategies, highlighted that Luna can complete tasks at one-fifth the cost of Sol while delivering "comparable performance" for many applications. This tiered pricing model directly addresses one of the primary barriers to enterprise AI adoption: the high cost of deploying advanced models at scale. However, questions remain about whether smaller models can sustain performance parity over time, leaving some industry experts cautious about long-term efficiency claims.

ChatGPT Work itself became available on July 9 to Pro, Enterprise, and Edu users via web and mobile platforms, with access expanding to Plus and Business tiers shortly afterward. The staggered rollout suggests OpenAI is prioritizing stability and security as it scales the product to a broader user base.

An intensifying agent war with Anthropic

ChatGPT Work enters a market where Anthropic’s Claude Cowork has already established a foothold. Launched in January 2026, Cowork initially targeted desktop users but expanded to mobile and web in July 2026, with Anthropic temporarily doubling usage limits through August 5, 2026. The timing of this expansion—just days before OpenAI’s announcement—suggests a strategic effort to preempt its rival’s entry into the autonomous agent space.

Ty Geri, OpenAI’s product manager for ChatGPT Work, emphasized the tool’s versatility across industries, particularly in automating workflows that traditionally required coding expertise. Unlike earlier AI assistants, which excelled at single-turn interactions, ChatGPT Work is designed for autonomous, multi-step execution—a capability Anthropic has already demonstrated with Cowork. The competitive landscape grew even more crowded on July 8, 2026, when Meta released Muse Spark 1.1, a coding model aimed at the same enterprise market.

While OpenAI has not disclosed specific pricing for the Sol, Terra, and Luna tiers, its efficiency claims imply a potential cost advantage. If Luna’s performance holds up under real-world conditions, it could force competitors to introduce similarly low-cost alternatives. For now, the focus remains on proving that smaller models can match the output of their larger counterparts without sacrificing reliability—a critical factor for enterprises evaluating long-term AI investments.

Security and geopolitics shape the AI landscape

The US government’s intervention in GPT-5.6’s release highlights a growing concern: advanced AI models may be capable of identifying and exploiting vulnerabilities in critical software systems. Similar apprehensions surround Anthropic’s Mythos series, which has also faced regulatory scrutiny. Policymakers fear that if such models are misused, they could automate cyberattacks or reverse-engineer proprietary code, posing risks to national security and economic stability.

Geopolitical tensions are further complicating the AI race. On June 25, 2026, Anthropic accused Alibaba of illicitly accessing Claude through fake accounts to conduct "distillation attacks"—a technique used to extract a model’s knowledge for training competing systems. The allegation underscores the intensifying US-China tech rivalry, particularly in AI development. Concurrently, Alibaba has filed a lawsuit against the US Pentagon to challenge its blacklisting over alleged ties to China’s People’s Liberation Army (PLA). These disputes signal a fragmenting global AI ecosystem, with US-based companies increasingly restricting foreign access to their models amid national security concerns.

For multinational corporations, these developments introduce new layers of compliance risk. If AI tools become subject to export controls or usage restrictions, businesses may need to regionalize their deployments or adopt localized alternatives. This trend is already visible in markets like China, where domestic models dominate due to limited access to Western AI systems. The regulatory environment is evolving rapidly, and companies must navigate these complexities to avoid disruptions in their AI-driven operations.

Enterprise adoption and the road ahead

The launch of ChatGPT Work and the rapid evolution of Claude Cowork mark a pivotal shift from experimental AI to mission-critical business tools. Enterprises are no longer merely testing generative AI; they are integrating it into core workflows, from legal document review to software development. OpenAI’s tiered pricing and efficiency claims could accelerate this adoption by making AI more accessible, while Anthropic’s early lead in autonomous agents may give it an advantage in user experience and reliability.

Both companies are also eyeing the public markets, adding another dimension to their rivalry. Reports suggest OpenAI may delay its IPO until 2027, but the pressure to monetize enterprise demand could expedite those plans. Anthropic, too, is reportedly preparing for a public offering, setting the stage for a high-stakes battle that will likely hinge on which platform can demonstrate the most tangible return on investment for businesses. The outcome of this competition could influence the broader AI industry, from pricing models to innovation cycles.

Yet, security and geopolitical challenges remain significant hurdles. If governments impose stricter controls on AI model releases—or if corporate users grow wary of compliance risks—the growth of these tools could be constrained. The US review of GPT-5.6 may be just the first of many such interventions, forcing AI developers to strike a delicate balance between innovation and risk mitigation. For global professionals, the message is clear: the era of AI-driven productivity has arrived, but its trajectory will be shaped as much by regulatory and geopolitical factors as by technological advancements.

As the rivalry between OpenAI and Anthropic intensifies, the next phase of the AI race will be defined not only by technical capabilities but also by which company can best navigate the complex interplay of cost, performance, and regulation. The stakes are high, and the outcomes will have far-reaching implications for businesses, governments, and the future of work itself.