Key Takeaways
- • Anthropic released Claude Opus 4.7 on April 16, 2026, positioning the model as a significant upgrade in coding, vision processing, and autonomous task execution, with improvements in handling long-running complex workflows and self-verification capabilities.
- • The company publicly acknowledged that Opus 4.7 underperforms relative to its unreleased Mythos model, which remains available only to a handpicked group of tech and cybersecurity companies due to safety concerns.
- • Opus 4.7 launch follows weeks of documented user complaints that Opus 4.6 had experienced performance degradation, with enterprise developers reporting the model "cannot be trusted to perform complex engineering," fueling speculation about deliberate "nerfing."
- • Anthropic attributed performance issues to token optimization rather than compute reallocation, with the company reducing default "effort" levels to economize on processing units—a cost-control measure that triggered user backlash and subscription cancellations.
- • The release intensifies competitive pressure in enterprise AI as Anthropic, recently valued at $380 billion and reportedly preparing for IPO, attempts to retain users amid rival offerings from OpenAI (ChatGPT 5.4) and Google (Gemini 3.1 Pro).
Summary
Anthropic released Claude Opus 4.7 on April 16, 2026, as a material upgrade to its flagship AI model designed to address enterprise demand for reliable autonomous task execution. The new model introduces enhanced coding capabilities, substantially improved vision processing (supporting up to 2,576 pixels on the long edge, approximately 3.75 megapixels—more than three times earlier versions), and a self-verification mechanism that allows the model to audit its own outputs before delivery. Anthropic also introduced a new "xhigh" reasoning effort level and a beta task-budget feature enabling organizations to manage token consumption across extended operations. According to Anthropic's announcement, the model handles long-running tasks with improved rigor and allows users to "hand off their hardest work with less supervision."
However, the release occurs amid significant user backlash regarding earlier performance declines. In mid-April 2026, developers and enterprise users widely reported that Opus 4.6 had regressed, failing to follow instructions reliably and making more errors on complex workflows. An AMD senior director's post on GitHub stating "Claude has regressed to the point it cannot be trusted to perform complex engineering" circulated broadly, catalyzing speculation that Anthropic had deliberately degraded the model—colloquially termed "nerfing"—either to manage computational costs or redirect resources toward frontier projects.
Anthropic's response acknowledged that recent changes had reduced the model's default "effort" level to economize on token processing, but the company denied that degradation was designed to redirect compute toward other initiatives. Employees, including one identified as Cherny, publicly pushed back against nerfing accusations on social media. Despite these denials, user sentiment deteriorated; social media commentary indicated subscription cancellations and reduced confidence in the platform's reliability for production workflows.
Anthropic's disclosure that Opus 4.7 trails its unreleased Mythos Preview model adds another layer of complexity. Mythos, distributed only to select tech and cybersecurity firms, reportedly outperforms Opus 4.7 across benchmarked tasks, yet remains unavailable to the broader market due to safety considerations. Opus 4.7 does exceed performance of competing models including OpenAI's ChatGPT 5.4 and Google's Gemini 3.1 Pro in published benchmarks, positioning Anthropic competitively in the enterprise AI race. The company's recent $380 billion valuation and reported preparation for public offering add urgency to resolving user trust and performance consistency concerns before the anticipated capital event.
However, the release occurs amid significant user backlash regarding earlier performance declines. In mid-April 2026, developers and enterprise users widely reported that Opus 4.6 had regressed, failing to follow instructions reliably and making more errors on complex workflows. An AMD senior director's post on GitHub stating "Claude has regressed to the point it cannot be trusted to perform complex engineering" circulated broadly, catalyzing speculation that Anthropic had deliberately degraded the model—colloquially termed "nerfing"—either to manage computational costs or redirect resources toward frontier projects.
Anthropic's response acknowledged that recent changes had reduced the model's default "effort" level to economize on token processing, but the company denied that degradation was designed to redirect compute toward other initiatives. Employees, including one identified as Cherny, publicly pushed back against nerfing accusations on social media. Despite these denials, user sentiment deteriorated; social media commentary indicated subscription cancellations and reduced confidence in the platform's reliability for production workflows.
Anthropic's disclosure that Opus 4.7 trails its unreleased Mythos Preview model adds another layer of complexity. Mythos, distributed only to select tech and cybersecurity firms, reportedly outperforms Opus 4.7 across benchmarked tasks, yet remains unavailable to the broader market due to safety considerations. Opus 4.7 does exceed performance of competing models including OpenAI's ChatGPT 5.4 and Google's Gemini 3.1 Pro in published benchmarks, positioning Anthropic competitively in the enterprise AI race. The company's recent $380 billion valuation and reported preparation for public offering add urgency to resolving user trust and performance consistency concerns before the anticipated capital event.
Go deeper than the headline
Research every firm in this story, track the ones that matter and get the daily briefing — free for 7 days.