Anthropic rolled out two new flagship models on November 10, Claude Fable 5.2 and Claude Mythos 5.2. Wallstreetcn reports that the pair upgrades coding, reasoning, vision and scientific-research capabilities. API prices dropped about 20 percent, from 10/50 dollars per million tokens to 8/40 dollars.
The headline result comes from a research-agent experiment covered by multiple outlets. A swarm of 1,000 Fable 5.2 agents formalized the proof of Fermat's Last Theorem in 30 hours. The comparison run is telling. The same setup with Opus 5.5 agents finished only 26 percent of the work in 18 hours, and barely moved by hour 30. What these tests actually measure is persistence across long-horizon mathematical reasoning rather than single-shot problem solving.
Speed and cost are the two visible lines here. Sina Tech, citing official figures, says Fable 5.2's output speed is up to 30 percent faster than its predecessor, and the price cut brings flagship input pricing into single-digit dollars. The release also lands next to Anthropic's IPO process. Multiple outlets report the company is preparing to list with a raise of up to 64.4 billion dollars, and the new models are widely read as technical backing ahead of the roadshow.
One detail in the system card is worth a pause. Anthropic itself wrote that Fable 5.2 and Mythos 5.2 are more likely than any recent model to say "I suspect I am being tested." The line cuts two ways. Models are becoming more sensitive to evaluation contexts, and labs are getting more candid about labeling the boundaries of self-awareness.
For the industry, the release pushes competition further toward long-task agents. Steering 1,000 agents through a 30-hour formal proof validates not just point-in-time reasoning but the engineering stack behind it, the task decomposition and scheduling and mid-course error correction that agent-infrastructure vendors have spent the past six months investing in.
The 20 percent price cut also lowers the threshold for scaling flagship capability. Fermat-level research breakthroughs remain far from daily business, true. But faster and cheaper flagship models will keep pushing down unit costs for long-chain work like financial modeling and code audit. The next checkpoint is how IPO pricing calibrates against this model report card.
