Anthropic launches Claude Opus 4.7, reclaiming top spot as most powerful LLM while

Anthropic’s Claude Opus 4.7 hit the market Tuesday, nudging ahead of OpenAI’s GPT‑5.4 and Google’s Gemini 3.1 Pro on several headline benchmarks, VentureBeat reported. The model posted an Elo rating of 1,753 on the GDPVal‑AA knowledge‑work evaluation, outpacing GPT‑5.4’s 1,674 and Gemini’s 1,314, marking the first time a publicly available LLM has reclaimed the top spot since OpenAI’s March release of GPT‑5.4. Anthropic said the gain comes from refinements in agentic coding, scaled tool‑use, agentic computer use and financial analysis, areas where the company has focused its recent research investments.

Despite the lead, the margin is razor‑thin. VentureBeat’s benchmark chart shows Opus 4.7 beating GPT‑5.4 by only 7‑4 points on directly comparable tests, underscoring how tightly contested the race has become. The new model also lags behind its rivals in specific domains: GPT‑5.4 still dominates agentic search (89.3 % vs. Opus 4.7’s 79.3 %) and retains an edge in multilingual Q&A and raw terminal‑based coding, according to the same source. Anthropic positioned Opus 4.7 as a “specialized powerhouse,” not a universal champion, acknowledging that the AI landscape remains fragmented across task categories.

Anthropic is simultaneously holding back its next‑generation system, Mythos, limiting access to a handful of enterprise partners for cybersecurity testing. Bloomberg detailed how internal stress‑tests revealed that Mythos could expose and exploit vulnerabilities in critical software stacks, prompting the company to deem the model “too dangerous” for broader release. The testing program, overseen by security researcher Nicholas Carlini, demonstrated that Mythos could rapidly identify and patch weaknesses, but also raised alarms about potential misuse for espionage or sabotage. As a result, Anthropic has kept Mythos under strict controls while leveraging its findings to harden Opus 4.7.

The strategic split between Opus 4.7 and Mythos reflects Anthropic’s dual‑track approach: push a market‑ready, high‑performance LLM while containing a more powerful, yet risk‑laden, successor. VentureBeat noted that the company continues to iterate on Mythos for a limited set of partners, focusing on defensive applications in banking and government sectors. Bloomberg added that banks and agencies are already evaluating the threat posed by Mythos, indicating that the model’s capabilities have attracted both commercial interest and regulatory scrutiny.

Industry observers see Anthropic’s move as a bid to regain momentum after OpenAI’s aggressive rollout of GPT‑5.4 earlier this year. The narrow lead on GDPVal‑AA suggests Anthropic can still compete on core knowledge‑work tasks, but the ongoing gaps in search and multilingual performance hint that the race will likely swing back and forth. With both OpenAI and Google continuing to release incremental upgrades, the next quarter could see another reshuffling of the leaderboard, making Opus 4.7’s reign potentially short‑lived.