Orient: Mythos reappears with refreshed vigor, as model ability steadily increases.

Orient released a research report stating that Anthropic officially launched the Mythos-class model, and the competition of large models is shifting from general question answering to high-intensity task execution. The capabilities of large models are still improving, and the AI industry chain is expected to benefit. The improvement in high-precision SVG, web game, and front-end generation capabilities indicates that models are transitioning from text code generation to design-intensive workflows. The main points of Orient are as follows: Anthropic officially launched the Mythos-class model, shifting the competition of large models from general question answering to high-intensity task execution. Anthropic recently released Claude Fable 5, defining it as a Mythos-class model for mass use; at the same time, the less restrictive Claude Mythos 5 is still open to some organizations through a trusted access mechanism. Claude Mythos Preview achieved approximately 52 times acceleration in the training code optimization task under clear experimental constraints, compared to about 4 times for skilled human researchers in a similar task. This demonstrates its significant advantage in code tasks that require clear structure, verifiability, and repeatable optimization, validating that the next generation of large models is evolving towards longer periods, more complex tasks, and higher risk capabilities. The report believes that coding capability is becoming a key direction for large model manufacturers to widen the gap, especially in tasks such as compiler optimization, parallelization restructuring, memory management, training framework optimization, and research automation. AI's advantages in large-scale search and rapid verification are expected to first manifest in efficiency improvements beyond traditional manual development. The enhancement of high-precision SVG, web game, and front-end generation capabilities indicates that models are transitioning from text code generation to design-intensive workflows. The Mythos-class model excels in complex SVG, web games, UI components, graphic generation, etc. Complex SVG generation is not just drawing, but also requires the model to understand spatial structure, lighting relationships, gradient effects, vector positioning, and hierarchical organization simultaneously, placing high demands on the model's fine reasoning and code expression capabilities. If the model can consistently generate high-quality, runnable, and modifiable front-end components, its impact will not be limited to developers writing code more efficiently, but may further reshape the traditional workflow from design drafts to front-end code. The report believes that the improvement in large model front-end generation capability will drive AI coding from backend logic, script generation to extending further into more visually and experientially focused scenarios such as web development, UI design, interactive prototyping, and game production, making competition among AI development tools such as Cursor, Claude Code, and Codex more intense. The recursive loop of AI participating in its own code production is forming, and security grades and high cost constraints are also rising simultaneously. Anthropic recently revealed that Claude has already been involved in generating a large amount of production code internally, and the proportion has rapidly increased from early 2025, showing that AI is transitioning from assisting engineers in writing code to participating in the production level code development, debugging, optimization, and research decision-making process. If subsequent Mythos 5 or similar models continue to break through in training code optimization, frontend engineering, multimodal generation, and complex task execution, large model manufacturers will have the opportunity to form a positive feedback loop of "AI writing codeoptimizing training systemsenhancing next-generation modelsfurther strengthening code capabilities." Meanwhile, the release of Mythos-class models also highlights the trend of hierarchical control of cutting-edge models in high-risk capabilities such as network security and biochemistry: the public version expands the available range through security restrictions, while stronger versions gradually open up through trusted access mechanisms. The report believes that in the next stage of large model competition, the focus will not only be on the expansion of parameter scale, but also on the efficiency improvement of models in real R&D processes, controllable release capabilities, reasoning cost control, and enterprise-level security compliance capabilities, driving continuous growth in AI development tools, code generation platforms, model inference infrastructure, and computing power requirements. Risk Warning Increased industry competition; Industry progress falls short of expectations.