What's this about?
Less than three months after Fable 5. The two models are the *same model* with different levels of safeguards: Fable 5.1 is generally available, Mythos 5.1 only through trusted access programmes with guardrails designed to support cybersecurity and life sciences work. Three things beyond raw capability: price drops about 25% for typical workloads via cheaper cache reads, a new Enterprise Frontier Safeguards system addresses two months of enterprise data-retention complaints, and — the line worth reading twice — "Fable 5.1 can now be used to discover software vulnerabilities, though not to develop exploits for them."
What does "same model, different safeguards" mean?
The weights are identical; what differs is the layer of refusal training and runtime filtering wrapped around them. This is a deliberate design: rather than train one model to be safe for everyone, which makes it useless for security researchers and biologists, they train the capability once and gate access to it. The trade is explicit — the restricted version exists, someone maintains the access list, and the difference between the two products is a policy decision rather than a technical one.
What's the community saying?
- The top comment is not about the model: "This time it came with a usage reset." The thread's centre of gravity is quotas.
- A specific complaint that generated real discussion: the reset pushed one user's 5-hour window from two hours away back to five, effectively costing them a reset in the weekly cycle.
- A workaround people share: schedule a cron job sending a trivial `claude -p` prompt via Haiku to align your usage windows where you want them.
- Substantive scepticism from an existing user: has anyone been able to get anything done with Fable in the first place? The alignment checks were sensitive enough that it kept pushing them back to Opus.
- Another names the specific irritations that drove them to GPT-5.6 Sol — laziness, load-bearing language, and a penchant for inventing dense conceptual names — while noting the announcement claims a less annoying writing style.