Anthropic launches Claude Opus 4.7, touts stronger advanced programming skills

Anthropic today released Claude Opus 4.7, its latest public AI model, with a clear pitch: hand it your hardest coding problems. IT Home (IT之家) has reported the update lands just two months after Opus 4.6, continuing Anthropic’s accelerated cadence of incremental releases.

What changed

Opus 4.7 focuses on high-end software engineering capabilities. Reportedly it handles ultra‑difficult, long-running tasks more reliably, understands complex instructions more precisely, and even designs internal verification steps before returning results. Anthropic says the model’s multimodal and visual abilities are improved as well — supporting higher‑resolution image recognition and producing more polished interfaces, slides and documents. Users have reportedly felt comfortable delegating previously human‑supervised, complex coding work to Opus 4.7.

It has been reported that Opus 4.7 also improves instruction‑following, multi‑session memory (using a file‑system based memory mechanism), and overall practical workplace skills — allowing it to remember key information across extended sessions and reduce repeated context input.

Benchmarks and token rules

According to the report, Opus 4.7 outperforms Opus 4.6, GPT‑5.4 and Google’s Gemini 3.1 Pro on a range of agent coding and computer‑operation benchmarks, but trails the more capable Claude Mythos Preview — which Anthropic has not broadly opened to the public. It has been reported that Mythos is currently only provided to select platform partners such as Apple. Note: these comparative claims are from third‑party reporting and have not been independently verified here.

Anthropic warns users that Opus 4.7 uses a new tokenizer and a more “thoughtful” high‑compute mode, so token consumption may rise — roughly 1.0–1.35x depending on content — and high‑compute interactions will produce more output tokens.

Availability and context

Anthropic also said it is formalizing a more predictable upgrade rhythm; Opus 4.7 follows two‑month gaps established since Opus 4.5. On the Mac client front, Claude Code’s automatic mode is now available to Max subscribers (previously limited to teams, enterprises and API users) and a new /ultrareview command launches a dedicated code‑review workflow.

This release comes amid growing geopolitical scrutiny of advanced AI models and platform partnerships. Who gets access — and under what commercial or regulatory terms — matters as much as raw capability. It has been reported that Anthropic is balancing broader public availability for Opus with more restricted rollouts of its top‑tier Mythos builds to strategic partners.