Claude Opus 4.7

Anthropic introduces Claude Opus 4.7, touting significant advancements in software engineering, vision, and handling complex tasks, alongside new cybersecurity safeguards. While the company highlights performance gains and unchanged pricing, the announcement triggers widespread user skepticism and frustration on Hacker News, fueled by perceived model degradations, increased token usage, and concerns about Anthropic's transparency and business practices.

584

Score

469

Comments

Highest Rank

on Front Page

First Seen

Apr 16, 3:00 PM

Last Seen

Apr 16, 8:00 PM

Rank Over Time

The Lowdown

Anthropic has officially launched Claude Opus 4.7, presenting it as a substantial upgrade to its predecessor, Opus 4.6. The new model boasts enhanced capabilities across several critical areas, aiming to improve user experience and efficiency.

Key features and improvements highlighted in the announcement include:

Advanced Software Engineering: Opus 4.7 is particularly noted for its rigor and consistency in tackling difficult coding tasks, devising self-verification methods, and precisely following instructions, reducing the need for close supervision.
Enhanced Vision and Multimodality: The model offers significantly better vision, processing higher-resolution images (up to 3.75 megapixels) for detailed visual tasks, and exhibits more refined taste in generating interfaces, slides, and documents.
Cybersecurity Focus: Following 'Project Glasswing,' Opus 4.7 incorporates new safeguards to detect and block high-risk cybersecurity requests, with a 'Cyber Verification Program' available for legitimate security professionals. This also implies that a more powerful 'Mythos Preview' model exists but remains under restricted release due to safety considerations.
New Control Features: A new xhigh effort level provides finer control over reasoning vs. latency, task budgets are in public beta, and Claude Code gains an /ultrareview command for in-depth code assessments.
Token Usage Considerations: Despite consistent pricing, an updated tokenizer means the same input can consume 1.0–1.35x more tokens, and higher effort levels lead to more output tokens. Users are advised to re-tune prompts and manage token usage, as Opus 4.7 follows instructions more literally than previous versions.

Anthropic positions Opus 4.7 as a robust step forward, emphasizing its improved performance on benchmarks and positive feedback from early testers, particularly for complex, long-running agentic workflows. However, user feedback regarding token usage and perceived 'nerfing' of prior models suggests a nuanced adoption landscape.

The Gossip

Degradation Dilemmas

Many users express deep skepticism about Opus 4.7's promised improvements, citing a consistent pattern of performance degradation and 'nerfing' of previous models, especially Opus 4.6, often coinciding with new releases. This has led to distrust, with some speculating that Anthropic intentionally reduces model quality to create a more dramatic 'upgrade' effect, or due to compute limitations. While Anthropic employees often deny degradation, users' anecdotal experiences paint a different picture, causing frustration and a feeling of being 'gaslit'.

Token Troubles and Pricing Pains

A significant concern revolves around token usage and its impact on subscription costs. Users are wary of the increased token consumption in Opus 4.7 (due to a new tokenizer and higher 'effort' levels), fearing it will rapidly exhaust their paid limits. Many criticize Anthropic's opaque communication regarding pricing structures and perceived 'token shrinkage,' leading to unpredictable expenses and a sense that their subscriptions offer less value. Some point to GitHub Copilot's increased per-prompt charge for Opus 4.7 as an indicator of rising underlying costs.

Cybersecurity Clampdown

The introduction of stricter cybersecurity safeguards in Opus 4.7, which block 'prohibited or high-risk' uses, has sparked outrage among security professionals. Users report the model refusing legitimate tasks like vulnerability research or penetration testing, even with explicit disclaimers about authorized defensive work. This move is seen as hindering vital security efforts and forcing users to consider less capable (or open-source) alternatives, highlighting a tension between safety mandates and practical application.

The Codex vs. Claude Conundrum

Many users actively compare Claude Opus with OpenAI's Codex (or GPT-5.4), with a notable trend of users migrating to Codex due to perceived superior reliability, consistency, and transparency. While some acknowledge Claude's strengths, particularly in its December 2025-era performance, others find Codex to be more 'self-correcting,' less 'sycophantic,' and better at achieving tangible results, despite potentially being 'sloppier' or having a less refined 'personality.' This intense competition underscores the volatility of user loyalty in the rapidly evolving LLM landscape.