ChatGPT Work.

Detailed Analysis

The article "ChatGPT Work" arrives as part of a weekly AI newsletter roundup, but its more revealing content lies in the surrounding news items, which sketch a landscape where Anthropic's Claude models remain the benchmark against which competitors measure themselves—even as OpenAI narrows the gap. The piece explicitly states that a year ago, "Claude Code is the only way to go," suggesting Anthropic held a commanding lead in coding-focused AI tools. The framing now, however, is that GPT-5.6 (referred to informally as "Sol") has become "genuinely better than Fable on many tasks," with "Fable" apparently serving as a colloquial or code-name reference to Claude models. This shift signals intensifying competition in the developer-tools and coding-assistant space, an area Anthropic has aggressively cultivated as its primary market differentiator since the release of Claude Code.

Despite the competitive pressure implied by ChatGPT's rise, several data points in the newsletter underscore Claude's continued relevance and strong positioning. Moonshot's newly released Kimi K3—a massive 2.8 trillion parameter open-weight model—is explicitly benchmarked against "Claude Fable 5" and GPT-5.6, with the article noting K3 "only trails" these two systems. This positions Claude's flagship model as one of only two systems in the world that a major open-weight competitor cannot surpass, a notable vote of confidence in Anthropic's frontier positioning even amid rapid commoditization from Chinese labs. More strikingly, the newsletter highlights a mathematical breakthrough attributed to Claude: "Claude Fable 5 produced a one-line formula that breaks the Jacobian conjecture," a problem unsolved since 1939, allegedly solved "with a single prompt." The detail that "even Terence Tao had to ask ChatGPT about it" is a telling aside—implying that Claude's output was significant enough to require cross-verification by one of the world's foremost mathematicians using a competing AI system. Whether fully accurate or exaggerated newsletter hyperbole, this anecdote reflects a broader narrative Anthropic has cultivated: Claude models as tools capable of genuine scientific and mathematical reasoning, not just conversational assistance.

Business decisions around pricing further illuminate Anthropic's strategic posture. The article notes that "Anthropic reversed Fable 5 pricing two days after free access expired," making the model "permanently free for all paid users." This kind of rapid reversal suggests Anthropic is highly sensitive to user backlash and competitive dynamics—likely responding to complaints or to competitive pressure from OpenAI's Work harness and Moonshot's aggressively priced Kimi K3 (priced at half of Opus 4.8's rate). Pricing wars among frontier labs have become a defining feature of the AI market in 2025-2026, as the marginal cost of inference drops and open-weight models from Chinese labs like Moonshot and DeepSeek apply downward pressure on closed-model pricing. Anthropic's willingness to adjust pricing quickly indicates it is treating user retention and goodwill as paramount, especially as Claude faces credible substitutes at lower price points.

Finally, the newsletter's mention that "Claude can now learn skills by watching you"—via a screen-recording feature that converts demonstrated tasks into reusable "skills"—points to a broader industry trend toward agentic, self-improving AI assistants that learn from user behavior rather than requiring explicit prompting or configuration every time. This feature, embedded in the desktop app's "+" menu, mirrors the article's own thesis about ChatGPT Work: that connected tools, persistent context, and computer-use capabilities constitute the next competitive frontier beyond raw model quality. As both Anthropic and OpenAI race to build "harnesses" that embed their models into users' actual workflows—spreadsheets, presentations, CRMs, browsers—the competition is shifting from benchmark performance alone toward who can best operationalize AI within existing enterprise and personal workflows. Claude's continued benchmark strength, combined with its parallel investment in agentic skill-learning, suggests Anthropic is racing on both fronts simultaneously, even as OpenAI's Work harness seemingly pulls ahead in integration breadth for now.