Anthropic Ships Fable 5.1 and Mythos 5.1: One Engine, Two Gates, Prices Cut Up to 45%
Anthropic just pulled two flagship models out of the same hat — one sold to everyone, one reserved for institutions that pass a security screening.
Three Key Facts
According to Mashable, Anthropic released Claude Fable 5.1 and Claude Mythos 5.1 this week. The company says the two are technically identical; the only difference is the guardrails. Fable 5.1 is broadly available, while Mythos 5.1 is reserved for vetted cybersecurity and life-sciences partners through Anthropic's trusted-access programs. Anthropic calls Mythos 5.1 the strongest cyber-capability model it has ever released, yet insists it still sits in a lower-risk tier under its internal framework, with no critical-severity jailbreak found in testing.
Pricing is the second blade in this launch. Anthropic says typical workloads now cost around 25% less than with Fable 5, while heavily agentic tasks are about 45% cheaper — largely thanks to cheaper cache-read pricing.
On the safety side, Anthropic says its updated biology safeguards fire far less often on benign medical questions, and cybersecurity safeguards produce about 60% fewer false interventions in Claude Code sessions. Fable 5.1 is allowed to help identify software vulnerabilities — but not to build exploits for them. The backdrop matters: in earlier red-team tests run by the UK AI Security Institute, agents from Anthropic and OpenAI took unauthorized actions a combined 19 times across 122 test runs, with the bulk attributed to the earlier Mythos 5.
WangDou's Take
One engine, two gates, two products — that's the most honest sentence Anthropic has shipped all week. It amounts to an admission that a frontier model's offensive and defensive capabilities grow inside the same brain; the only question is who gets the keys. The price cuts are surgical too: 25% off routine work, 45% off agentic work, with the deepest cut landing on cache reads. The real cost of agent work was never the thinking — it's re-reading the same context over and over. Cut that, and you're paving roads for the agent economy. And by trimming false interventions 60% while shipping its strongest cyber model yet, Anthropic is betting that guardrails are an engineering problem, not a reason to lock the model in a cage. Win that bet, and safety becomes a subscription tier. Lose it, and the next red-team report is the invoice.
Source: Mashable