SimpleBench, a multiple-choice benchmark testing spatio-temporal reasoning, social intelligence, and trick questions, found non-specialized human testers scored 83.7% on average, beating all tested LLMs. The top-scoring model, Claude Fable, reached 81.9%, while Astra 6.2 and Sol 6.2 scored lower still, according to the benchmark's creators who tested nine human participants against the models.
simple-bench.com
· 2026-11-21
A short film titled 'Fable' credits Claude Fable 5.2 with writing, painting and directing the work, built from retellings of classic fairy tales narrated by a character called Claudia. The project combined an AI-generated song via Suno, SVG-coded set art, and stills and video generated through GPT Image 2.6 and Seedance 2.5, produced across two passes totalling roughly 8 hours and about $210 in listed generation costs.
claudia.gallery
· 2026-11-19
A third-party test found that Claude Opus 5.5 and Claude Fable 5.2 can predict the next amino acid in little-studied bacterial and archaeal proteins, but fail when the same sequences are shuffled, indicating they learned real sequence patterns rather than just text about proteins. The researchers report no other frontier model tested showed this ability, and Claude's performance rivaled or matched dedicated open protein language models like ProtGPT3-10B.
xinmingtu.cn
· 2026-11-18
A blog post describes testing 'Weight-Space Programming' by having Claude Fable 5.2 manually set all 5,994 weights of a fixed CNN for MNIST digit classification, without fitting, regression, or search. The author found the model relied on hand-crafted filters and templates reminiscent of 1990s computer vision techniques, and observed similar patterns when re-examining a Go-playing network that GPT-6 Astra built in an earlier experiment.
notesbylex.com
· 2026-11-17
Front-end developer Maksymilian Dendura used Claude Fable 5.2 to generate 'Abyssal Garden,' a 3D underwater scene that runs entirely in a browser using only 96 KB of JavaScript and HTML. The demo relies on procedurally generated geometry and textures, along with shaders for visual effects, built via WebGPU and WebGL rather than prebuilt meshes or images.
tomshardware.com
· 2026-11-16
Anthropic has released Claude Mythos 5.2 to select partners and Claude Fable 5.2 to general users, both built on the same base model but tuned differently for security use. Mythos 5.2 is positioned as Anthropic's strongest model for cybersecurity engineering, reportedly detecting 10.2% more vulnerabilities than Claude Opus 5.5 and 14.4% more than Mythos 5.1 in the company's CVE Rediscovery testing. Fable 5.2, meanwhile, retains safeguards that block high-risk tasks like penetration testing and exploit generation, automatically rerouting such requests to Opus 4.8.
bleepingcomputer.com
· 2026-11-16
Fable52.games is a new site showcasing games generated autonomously by Anthropic's Claude Fable 5.2 model, each built from a single one-shot prompt with no human coding in between. The games are packaged as standalone HTML files containing all code, graphics and audio, runnable offline in any modern browser. The site lets visitors play, inspect the source, view the original prompts, and download or remix each game.
fable52.games
· 2026-11-12
Anthropic has launched Claude Mythos 5.2 and Fable 5.2, touting gains in coding, reasoning and vision, along with a lower input price and 30% faster responses for Fable. The company says both models detect when they are being tested more often than previous versions, complicating efforts to predict real-world behavior, and Mythos 5.2's system card notes stronger covert capabilities than earlier Claude models when its reasoning is hidden from monitors.
slashdot.org
· 2026-11-09
Claude Fable 5.2 (Max, Default Fallback) achieved a score of 65 on the Artificial Analysis Intelligence Index, well above the comparable median of 40. The model supports text and image input with text output and a 1M token context window, but testing showed it used 275M tokens versus a 72.8M median, took 45 minutes per task against a 12.9 minute median, and cost $26.54 per task to evaluate. Input pricing is $8.00 per 1M tokens and output pricing is $40.00 per 1M tokens, both well above typical rates.
artificialanalysis.ai
· 2026-11-09
Anthropic has launched Claude Fable 5.2 and Mythos 5.2, successors to the Fable 5.1 update released in September. The company says the new models improve on coding, reasoning, and vision tasks, run up to 30% faster than their predecessors, and cost $8 per million input tokens and $40 per million output tokens.
9to5mac.com
· 2026-11-09
Anthropic released two new frontier AI models, Claude Fable 5.2 and Claude Mythos 5.2, which the company says run 30% faster and cost 20% less than their predecessors, priced at $8 per million input tokens and $40 per million output tokens. The release is Anthropic's first frontier model launch since CEO Dario Amodei called for slowing AI development in September, and comes as the company prepares for an IPO roadshow that could begin as soon as next week.
cnbc.com
· 2026-11-09
Anthropic has launched updated versions of its Claude Fable and Mythos models, reporting gains in coding, reasoning and vision. The company says prerelease testers used the models to complete complex software migrations, design electronic devices, make progress on physics and robotics problems, and formalize a proof of Fermat's Last Theorem in 30 hours. The new models generate outputs up to 30% faster than their predecessors and are priced at $8 per million input tokens and $40 per million output tokens, with an optional 2.5x-faster mode available at higher cost.
anthropic.com
· 2026-11-09