Anthropic Releases Claude Fable 5.2 and Mythos 5.2 for Coding and Science
Anthropic has introduced Claude Fable 5.2 and Mythos 5.2, bringing major performance gains in scientific reasoning, complex coding tasks, and multi-agent workflows at lower token prices.
Anthropic has officially introduced Claude Fable 5.2 and Claude Mythos 5.2, marking its most capable large language models to date. While both systems share the same core capabilities, they differ in their safety configurations. Fable 5.2 restricts specific biological and cybersecurity use cases, whereas Mythos 5.2 features relaxed guardrails and is offered exclusively to vetted researchers through two limited-access programs.
Advancing scientific and mathematical problem-solving
In testing across 14 artificial intelligence benchmarks, Fable 5.2 exceeded the performance of Fable 5.1 by more than 10% in most evaluations. The model demonstrated substantial strength in scientific domains, achieving a score of 74.6% on the Terminal-Bench-Science 0.1 benchmark, a 22% improvement over Fable 5.1.
Anthropic highlighted the model's problem-solving capacity across multiple complex theoretical challenges. Earlier in the year, the company utilized an unreleased model to build a computer-verifiable version of a well-known mathematical proof, a project that required 11 days. Fable 5.2 reduced that execution time to 30 hours by orchestrating 1,000 agents in parallel. In another evaluation, the model answered two open research questions regarding a theoretical physics concept linked to black holes.
Research capabilities also surfaced in tests of Mythos 5.2. When tasked with analyzing the protective shells, or capsids, of harmless adeno-associated viruses (AAVs) used in drug delivery therapies, Mythos 5.2 successfully modeled 28 out of 34 AAV capsids. Because capsids consist of dozens of distinct proteins, this is an intricate task. By contrast, specialized biology systems such as Google DeepMind's AlphaFold2—whose developers received half of the 2024 Nobel Prize in chemistry—are designed to model the structure of individual proteins rather than complete multi-protein complexes.
Guardrail routing and system-level coding
Anthropic also altered how Fable 5.2 enforces safety boundaries. Rather than outright refusal, prompts involving research biology or cybersecurity are automatically redirected to Opus 4.8 or Opus 5. The company stated that Fable 5.2 is 55% less likely to erroneously flag and block benign user requests compared to Fable 5.1.
Software engineering capabilities saw notable upgrades as well. Fable 5.2 registered improvements of more than 13% over Fable 5.1 on both the FrontierCode 2.0 and FrontierSWE 2.0 coding benchmarks. In an internal test, an Anthropic engineer tasked Fable 5.2 with creating an entire operating system. Deploying 1,000 parallel agents, the model finished the project in roughly eight days, producing an operating system that boots on a laptop, supports network connectivity, and features a built-in Claude sidebar.
What it means for developers
For engineering teams, Fable 5.2 and Mythos 5.2 provide specialized engines for long-horizon, open-ended technical challenges. Anthropic positions the models alongside Opus 5.5, which debuted in July for standard day-to-day automation. Teams can implement multi-tier setups where Opus 5.5 handles routine workloads and escalates complex blockers to Fable 5.2 or Mythos 5.2.
The new models also arrive with lower operational costs. Pricing is set at $8 per million input tokens and $40 per million output tokens, a 20% discount compared to previous-generation models. Additionally, both systems offer an optional high-performance execution tier that delivers prompt completions 2.5 times faster for twice the token price.
Developers evaluating model options can access top AI models cheaply through one API at https://apixoai.online, making it straightforward to test different architectures against technical requirements. With improved agent parallelization, fewer false-positive safety triggers, and expanded scientific reasoning, Anthropic's 5.2 lineup offers substantial new utility for advanced research and software development.
Source: Anthropic launches Fable 5.2 and Mythos 5.2 with science, coding upgrades — SiliconANGLE AI. Written by the Apixo team from that report.
One key for Claude, GPT, GLM, DeepSeek and more. Pay per token with crypto.
Get your API key
