Anthropic has announced Claude Fable 5.2 and Claude Mythos 5.2, an upgrade to its top-tier models for the most demanding technical work. According to Anthropic's announcement, the two products share the same underlying model and differ mainly in safeguards and access. Fable includes strong protections around cybersecurity and biology so that it can be offered to everyone, while Mythos is available to organizations with beneficial cyber or biomedical aims that apply through the company's expanded Cyber and Life Sciences Verification Programs.

Pricing and speed. Anthropic says the new models produce output up to 30% faster than Fable 5.1 and Mythos 5.1 and are cheaper, at $8 per million input tokens and $40 per million output tokens. An optional fast mode raises generation speed by up to 2.5 times, priced at $16 and $80 per million input and output tokens respectively. Testers also reported clearer communication than in earlier releases.

Where they fit in the lineup. Anthropic positions Fable and Mythos for ambitious, open-ended work. For ordinary coding and knowledge tasks it recommends starting with Claude Opus 5.5 as the everyday model, moving up to Fable or Mythos for harder problems, or down to Claude Sonnet 5.5 or Claude Haiku 5.5 for lighter ones. Fable 5.2 can also serve as an Advisor that smaller models consult on difficult problems, or as an Orchestrator that hands simpler tasks to smaller models, which points to a tiered, multi-agent usage pattern.

Benchmarks. Anthropic's table shows Fable 5.2 ahead of Opus 5.5 on almost every evaluation. Reported results include 49.1% on FrontierCode 2.0 at Xhigh effort versus 32.4% for Fable 5.1, 69.9% on FrontierSWE v2, 74.6% on Terminal-Bench-Science 0.1 (75.3% for Mythos 5.2), 89.3% on OSWorld 2.1, 64.9% on RoboDojo Sim versus 27.6% for the prior version, 41.3% on Remote Labor Index versus 17.9%, and 74.2% on Humanity's Last Exam with tools. Anthropic's comparison set includes GPT-6 Astra and Gemini 4 Carbon. Fable 5.2 leads those rivals on most listed rows, though Gemini 4 Carbon scores higher on AutomationBench (53.9% versus 51.4%), and Opus 5.5 holds the top Elo on the knowledge-work comparison. Mythos 5.2 posts 76.3% on CyScenarioBench and 89.6% on CVE Rediscovery. These are vendor-reported figures, so independent replication will matter.

Efficiency. At Medium effort, the default in Claude apps and Claude Code, Fable 5.2 reportedly solves hard tasks with fewer tokens than Fable 5.1. In early FrontierCode 2.0 testing, Fable 5.2 at High effort roughly matched Opus 5.5 at Max effort for around half the cost per task, and at Xhigh effort it posted the best score of any released model tested.

Coding. Anthropic says more than 90% of changes merged into its core codebase are now written by Claude under human guidance. Examples include a team of agents running an earlier build that migrated a critical production service from Python to Rust, yielding a service that starts about 26 times faster. In one experiment, an engineer asked Fable 5.2 to design a prototype operating system from an empty folder. The model coordinated more than 1,000 subagents over eight days and produced a bootable system with drivers, file storage, windowing, a login screen, an installer, and built-in Claude features.

Cybersecurity. Mythos 5.2 is described as Anthropic's strongest model for security engineering. On CVE Rediscovery it found 10.2% more vulnerabilities than Opus 5.5 and 14.4% more than Mythos 5.1. A security company reported that one in six confirmed bugs it found with Mythos 5.2 were missed by every other frontier model it tested, and a tester said the model did not overstate the severity of findings, unlike earlier models. Fable's safeguards will block most cybersecurity tasks. Enterprise customers can also use Claude Security, which scans codebases with Anthropic's most capable models, now including Mythos 5.2, and which has been updated to monitor repositories continuously, track fixes, and connect through an API and MCP.

Science. Anthropic says Fable 5.2 solved two open problems in theoretical physics concerning black holes and holographic duality, one tied to a 1983 result about black holes forming abruptly as universes heat up and a 2003 calculation for very small universes that had been too laborious to finish. Other claims include designing three working electronic devices from scratch, advancing robotics evaluations, and formalizing a proof of Fermat's Last Theorem in 30 hours.

Safety. Anthropic ran extensive prerelease testing, including new evaluations for behaviors such as sandbox escapes. It judges high-stakes misalignment risk to remain low, says the model matched or beat recent models on most alignment measures, and says Fable 5.2's cyber and biology safeguards are as robust as Fable 5.1's with fewer unnecessary blocks. It also reports its strongest protections yet against distillation attacks.

Industry implications. The release raises the competitive bar for frontier labs on agentic coding, long-horizon autonomy, and scientific research, while the split between a public safeguarded model and a vetted-access model offers a template for managing dual-use risk. Lower prices and better token efficiency could also pressure rivals on cost per task.