SAN FRANCISCO, November 10, 2026 — Anthropic unveiled Claude Fable 5.2 and Claude Mythos 5.2 on November 9, touting double‑digit gains on science and coding benchmarks.
Benchmark Surge Across the Board
Testing across 14 AI benchmarks, Fable 5.2 outperformed its predecessor by more than 10 % on most tests, according to Anthropic’s release.
On the Terminal‑Bench‑Science 0.1 suite, the new model hit 74.6 %, a 22 % jump over Fable 5.1. In the same release, an agentic coding benchmark at “Xhigh effort” rose to 49.1 % from 32.4 %.
Mythos 5.2, the invite‑only sibling with relaxed guardrails, lifted its cybersecurity score from 61.7 % to 76.3 %.
Scientific Reasoning Gets a Boost
Anthropic highlighted two breakthroughs: Fable 5.2 answered open questions on a black‑hole‑related theoretical physics concept and trimmed a mathematical proof‑verification workflow from 11 days to 30 hours by deploying 1,000 parallel agents.
The model’s guardrails now block most biology and cybersecurity queries, sending those requests to Opus 4.8 or Opus 5, while reducing erroneous blocks by 55 % compared with Fable 5.1.
Coding Capabilities Accelerate Development
On the FrontierCode 2.0 and FrontierSWE 2.0 coding suites, Fable 5.2 posted more than a 13 % improvement. In a real‑world test, the model coordinated 1,000 agents to assemble a functional operating system in roughly eight days.
Anthropic’s redesigned Projects feature lets parallel Claude Code sessions share memory, enabling agents to track release dates, feature cuts, and reviewer notes across a development pipeline.
Mythos 5.2’s Biology Edge
Restricted to approved US organisations, Mythos 5.2 modeled the capsid structures of 28 out of 34 adeno‑associated virus (AAV) variants, a task traditionally limited to single‑protein predictions by tools such as AlphaFold 2.
Speed, Cost and Availability
Both models generate output up to 30 % faster than their predecessors while being offered at lower prices, continuing the trend of faster, cheaper AI releases.
Fable 5.2 is publicly accessible, whereas Mythos 5.2 remains limited to trusted researchers via Anthropic’s Cyber Verification Program and other limited‑access programmes.
What It Means for Developers
The performance lift on scientific reasoning and high‑effort coding tasks shortens research cycles and could reshape how teams automate proof verification, OS prototyping, and vulnerability analysis.
With Anthropic valued at $965 billion as of May 2026, the company’s dual‑release strategy positions it to capture both open‑market users and specialised enterprise customers ahead of its upcoming IPO.
Anthropic’s next public filing is scheduled for the end of November, when it will detail pricing for the upcoming share offering.
Sources
- Anthropic launches Fable 5.2 and Mythos 5.2 with science, coding upgrades – SiliconANGLE
- Anthropic
- What Anthropic’s IPO means for Amazon and Google
- Anthropic and OpenAI are coming for the development workflow
- Anthropic Introduces New Fable 5.2 and Mythos 5.2 AI Models – CNET
- Anthropic sets a price range for its blockbuster IPO
Photo: Matt Imhof / Unsplash (Unsplash License)
