A major leak from Anthropic has revealed stunning details about their next-generation AI model, codenamed 'Mythos.' The system is rumored to feature an unprecedented 1 million token context window, allowing it to process information equivalent to over 700 novels at once. Internal evaluations suggest Mythos achieves over 90% accuracy on GPQA Diamond, the most challenging PhD-level science benchmark.
Unlike typical chatbots, Mythos is designed as an 'agent OS,' capable of autonomous operation on complex tasks for extended periods. Anthropic is reportedly prioritizing safety, delaying public release until comprehensive evaluations are complete, with a target of Q3 2026.
If these benchmarks hold, Mythos could represent the most significant leap in AI capabilities, creating the widest performance gap ever seen in the industry. Stay tuned for updates.