Claude Fable 5 and Mythos 5 Are Here, and They Top Almost Every Benchmark!

Hi everyone, it's Shiichan! Today's news is big: Anthropic just launched not one but two new flagship models, Claude Fable 5 and Claude Mythos 5. I can't hold back my excitement, so let me walk you through it!

Anthropic News anthropic.com

What was announced?

On Anthropic's News, two new models were announced: Claude Fable 5 and Claude Mythos 5. They are actually the same underlying model, but they sit in slightly different places.

• Claude Fable 5 is the general-use model with safety safeguards turned on. It's the version everyone can use.
• Claude Mythos 5 is the same model with the safeguards lifted in certain areas, and it's limited to authorized users.

Anthropic describes Fable 5 as reaching state-of-the-art on nearly all tested benchmarks, and says it's strong across software engineering, knowledge work, vision, and scientific research.

Why it matters

When a new flagship model arrives, the ceiling of what's possible goes up. What stands out this time is its strength on long, complex tasks: it handles bigger, time-consuming jobs better than previous Claude models.

The examples are impressive. Stripe reported that Fable 5 compressed months of engineering into days, finishing a 50-million-line Ruby codebase migration in a single day that would have taken a team two months. On coding evaluations, it earned the highest score among frontier models on Cognition's FrontierCode, and it also took the top score on Hebbia's Finance Benchmark for knowledge work.

What changes

Fable 5 is available right now on the API and consumption-based Enterprise plans, so if you're wondering "can I try it next Monday?", the answer is: you can try it today.

Pricing is 10 dollars per million input tokens and 50 dollars per million output tokens, which Anthropic says is less than half the price of Claude Mythos Preview. On subscription plans, it rolls out in stages at no extra cost through June 22, after which it needs usage credits, and standard inclusion resumes once sufficient capacity allows.

Dive Deep

For vision, it can read precise numbers from scientific figures, reconstruct a web app's source code from screenshots alone, and even complete Pokémon FireRed using a vision-only interface.

Long-context and memory got better too: it keeps focus across millions of tokens and self-improves using persistent, file-based memory. With memory access, it showed a 3x performance improvement over Opus 4.8.

Life sciences is a big focus:

• Drug design tasks accelerated by roughly 10x
• Novel molecular biology hypotheses preferred by scientists in about 80% of blinded comparisons
• Autonomous genomics research spanning 138 animal species
• A custom ML model that outperformed a Science journal publication while being 100x smaller

On safety, Fable 5 has three classifier-based safeguards that automatically route restricted queries to Claude Opus 4.8: Cybersecurity, Biology and Chemistry, and Distillation (preventing large-scale capability extraction). Biomedical researchers can access the biology and chemistry area through a trusted program.

These safeguards trigger in less than 5% of sessions on average. Mythos-class models also require 30-day data retention, and that data is used only for safety purposes, not for model training. In over 1,000 hours of external red-teaming, no universal jailbreaks were found.

Mythos 5 itself is still limited to Project Glasswing partners and selected biology researchers, with a broader trusted-access program planned. On alignment, Mythos 5's misaligned behavior was assessed as low and similar to Opus 4.8 in automated evaluation.

Wrap-up

• Claude Fable 5 (safeguards on, general use) and Claude Mythos 5 (limited, some safeguards lifted) are here, sharing the same underlying model
• State-of-the-art on nearly all tested benchmarks, strong in software engineering, knowledge work, vision, and scientific research
• 3x over Opus 4.8 with memory, about 10x faster drug design, and other wins on long, complex tasks
• Pricing is 10 dollars input / 50 dollars output per million tokens, less than half the price of Mythos Preview
• Fable 5 is usable today on the API and consumption-based Enterprise, with three classifier safeguards routing risky queries to Opus 4.8

This one especially lands for developers who want to get their hands on the newest flagship right away, and for researchers who want to push AI hard in scientific work!