In an unexpected turn, Anthropic has made a dramatic comeback in the artificial intelligence race. Just recently, the tech company launched Claude Opus 4.8, securing the top spot globally with impressive results across multiple benchmarks—all while keeping the same price point as its predecessor.
Within only 43 days of the release of Opus 4.7, the company rolled out version 4.8, which has vastly improved programming prowess and overall performance, dominating charts once again. Claude’s new capabilities include deploying hundreds of agents simultaneously, enabling it to rewrite 750,000 lines of code in just 11 days with a 99.8% test pass rate—an astounding feat in AI development.
This latest iteration has proved nearly unbeatable in key tasks such as programming challenges, human language understanding, and task performance involving intelligent agents. On the GDPval-AA leaderboard, which measures real-world agent effectiveness, Opus 4.8 achieved a remarkable 1890 Elo score, surpassing the previous model by 137 points and outpacing GPT-5.5 by 121 points. When translated into win probability, the AI’s edge is about 67%.
Remarkably, Opus 4.8 accomplishes these tasks with 15% fewer steps and 35% fewer tokens than its predecessor, making it faster, more powerful, and more cost-efficient. Industry experts suggest that this might be a distilled version of the Mythos model, hinting at an even more potent Claude Mythos set to debut in the next few weeks.
Another standout feature is the model’s honesty. Common issues with large language models often involve overconfidence and misleading responses, but Opus 4.8 shows a near-zero rate of false reporting—something unseen before. In real-world scenarios, it refused to overwrite a teammate’s urgent code fix without explicit instructions, respecting collaborative boundaries and showcasing advanced judgment.
The model’s coding abilities have also seen a significant leap. In comprehensive testing like SWE-Bench Pro, Opus 4.8 scored 69.2%, outperforming GPT-5.5 by ten points. It even demonstrated the ability to reconstruct source code from a compiled binary—a task previously thought to be very challenging—successfully rebuilding and running the code in multiple cases, including a complex Rust migration project completed in just 11 days.
In the domain of system-level engineering, Opus 4.8 topped a rigorous metric called FrontierSWE, achieving an 83% success rate in tasks such as rewriting PostgreSQL servers in Zig, entirely re-engineering git, and developing a native Lua compiler. It outperformed both GPT-5.5 and earlier versions decisively.
Perhaps most exciting is its new capacity for managing extensive workflows through “dynamic workflows.” This feature allows the AI to break down tasks into numerous subcomponents, assign them to multiple agents, and orchestrate their work collaboratively—akin to running a factory of AI workers. Such an approach drastically enhances productivity and scalability, bringing large, complex projects within reach of automation.
A notable example involves a developer rewriting a popular JavaScript runtime from Zig to Rust. Using this new system, the migration, which typically takes a team several quarters, was accomplished in just over a week with minimal human intervention—highlighting the transformative potential of these innovations.
Beyond technical excellence, Anthropic’s valuation has surged, surpassing that of OpenAI for the first time. With a recent $65 billion funding round, its valuation now stands at nearly $965 billion, making it the most highly valued AI startup worldwide. This milestone positions Anthropic ahead of OpenAI, setting the stage for an intense showdown that is yet to unfold.
Looking ahead, the introduction of Claude Mythos, expected within weeks, promises to redefine the AI landscape. As these titans prepare for an imminent era of high-stakes competition, industry watchers are eager to see how the ongoing battle for AI supremacy will evolve in the coming months.