Anthropic Just Dropped Two New Claudes. Here's What's Actually Different.
Anthropic just shipped two new models: Claude Fable 5 and Claude Mythos 5. Both go live today. Both are cheaper than what came before. And both are, by most benchmarks, the best models the company has ever released.
Here's what you actually need to know.
Two Models, Two Jobs
The short version: Fable 5 is the everyday workhorse, Mythos 5 is for high-stakes scientific and technical work.
Fable 5 is positioned as a general-purpose model with standard safety guardrails. It hits top scores on software engineering benchmarks, knowledge work tasks, vision, and research. Stripe says it compressed what would have been months of engineering into days. On FrontierCode — a benchmark for coding ability among frontier models — it ranks first.
Mythos 5 is the restricted-access model. It's been tested specifically in drug design workflows, where it delivers roughly a 10x speedup. In blind molecular biology hypothesis tests, scientists preferred its outputs about 80% of the time. That's not a small number.
The Price Cut Is Real
Both models are priced at $10 per million input tokens and $50 per million output tokens.
That's more than a 50% reduction compared to the Claude Mythos Preview pricing. For developers running high-volume workloads, that's the kind of change that meaningfully affects what's practical to build.
What Fable 5 Can Do That's Actually Impressive
One standout capability: Fable 5 can reconstruct a web application's source code from a screenshot alone. You hand it an image of a UI, and it writes the code. No spec, no existing codebase to reference — just the visual output.
That's genuinely useful for prototyping, competitive analysis, or porting designs. It's the kind of task that would have taken a skilled engineer hours.
On safety: Fable 5 does fall back to Opus 4.8-level responses on some sensitive topics, but Anthropic says this only affects fewer than 5% of sessions in practice. Most users won't notice it.
Mythos 5 and Project Glasswing
Mythos 5 isn't open to everyone. It's the more restricted of the two, aimed at research and science applications where the model's capabilities carry real-world consequences.
One notable access path: cybersecurity defenders can get access through Project Glasswing, Anthropic's initiative to put powerful AI in the hands of people working on defensive security. It's a deliberate choice about who gets access to the most capable tools first.
What This Means If You Use OpenClaw
OpenClaw runs on Claude — which means the models Anthropic ships today are the models your agent can use tomorrow.
When Anthropic improves Claude's ability to reason about code, handle complex tasks, and work across longer contexts, that flows directly into what your OpenClaw agent can do. The 10x speedup in scientific workflows isn't just useful for drug researchers — it's a signal that the underlying model is getting substantially better at sustained, multi-step work. That's exactly what an AI agent does.
If you're building automations on ClawWorld, today's releases mean your agent is working with more capable foundations — at a lower cost per token than before. That combination tends to unlock workflows that weren't practical six months ago.