LLM-Anthropic 0.31 (simonwillison.net)

Anthropic has unveiled its latest models, Fable 5.2 and Mythos 5.2, with Fable 5.2 currently topping the Artificial Analysis rankings. The company claims that these new models have made significant advancements by addressing "two open problems in the physics of black holes," indicating a leap in their internal research benchmarks and pushing the boundaries of AI applications in complex scientific domains. Fable 5.2 is now available for users to experiment with, enabling prompt-based interactions to generate creative outputs, such as an SVG of a pelican riding a bicycle. Notably, users experienced some limitations, with the highest reasoning level, "max," failing to return results due to running out of tokens—a bug not previously encountered in prior versions. Priced competitively at $8 per million input tokens and $40 for output tokens (a 20% discount from Fable 5.1), Fable 5.2 presents an accessible option for researchers and developers looking to leverage advanced AI for creative as well as technical problem-solving tasks.