CONFIRMED

Anthropic has launched Claude Fable 5.1 and Claude Mythos 5.1, direct successors to Fable 5 and Mythos 5. Per the company's official announcement, the two are the same underlying model shipped with two different sets of safeguards: Fable 5.1 is generally available to everyone, on Claude.ai, Claude Code, Claude Cowork and the API, while Mythos 5.1 remains restricted to trusted access programs aimed at cybersecurity and life sciences. Independent coverage from outlets such as Tech Insider and Thurrott places the launch on September 1, 2026.

An aggressive cut to cache pricing, everything else stays put

The detail drawing the most attention in the announcement is the cache read price: it dropped from $1 to $0.25 per million tokens, a 75% cut. Input and output pricing stays the same as Fable 5, at $10 per million input tokens and $50 per million output tokens. According to Anthropic, this drops the cost of a typical workload by roughly 25%, and the cost of a heavily agentic workload, one that reprocesses the same context many times, by as much as 45%. In practice, teams running long running coding agents that repeatedly re read the same context feel the cut far more than users asking one off questions.

Fewer false positives, and a new door for vulnerability hunting

Anthropic says it cut cybersecurity safeguard false positives per Claude Code session by 60%, and biology safeguard false positives on basic biology and medical questions by 85%. The most concrete change: Fable 5.1 can now be used to discover software vulnerabilities, something the previous version refused, though it still cannot develop exploits from them. External penetration testing by Trajectory Labs, with more than 6,500 requests over 74 hours, found no complete working exploit and no universal jailbreak, according to specialized press coverage.

Advanced biology access, only with US government sign off

For Mythos 5.1, Anthropic opened an access program for advanced biology capability developed in partnership with the US government, currently limited to a group of American organizations, with plans to expand to domestic and international partners. It's a direct continuation of the July episode, when the US government temporarily suspended Fable 5 and Mythos 5 over export control concerns, and Anthropic responded by tightening the vetting process rather than opening up free access.

Performance gains come with customer receipts

In benchmarks published by Anthropic, Fable 5.1 jumped from 24.7% to 52.6% on Terminal-Bench-Science 0.1, went from 42% to 55.8% on Terminal-Bench 4.0 (60.9% on the Mythos version), and beat Fable 5 on CursorBench 3.2.0, from 70.5% to 73.4%. Companies that tested the model ahead of public launch confirm the gain outside the lab: Cognition announced it will move traffic currently running on Opus 5 inside Devin over to Fable 5.1, citing lower cost per task. At Millennium, the model found the cause of a rare failure in internal systems that the engineering team had been unable to explain for four to five years.

Why it matters for automation teams in Brazil

For agencies and product teams running long running coding or research agents, the 75% cut in cache pricing changes the math on automations that today fall outside budget because they reprocess context repeatedly. It's the second time in a few months that Anthropic has adjusted pricing on its top tier model under direct customer pressure, after already canceling a Sonnet 5 price hike in August. The item still worth watching is Enterprise Frontier Safeguards, the new customer custodied monitoring scheme mentioned in the Fable 5.1 launch but still without a single confirmed general availability date, expected to arrive in phases this Northern Hemisphere fall.

MaxAssistant's read

The launch confirms a pattern: Anthropic is treating cache pricing, not raw token pricing, as this cycle's competitive lever, because that's where the real cost sits for anyone running a long running agent. It's a direct response to pressure from competitors like GPT 5.6 Sol and Gemini, which compete precisely in that kind of workload. The catch sits in the most capable part of the package: Mythos 5.1 remains rare access under government sign off, keeping the cybersecurity and biology frontier out of reach for the vast majority of teams that just want a cheaper agent to run.

Sources

Anthropic, Introducing Claude Fable 5.1 and Claude Mythos 5.1: https://www.anthropic.com/claude-fable-and-mythos-5-1 | Tech Insider, Anthropic Fable 5.1, Mythos 5.1 Cut Cache Cost 75%: https://tech-insider.org/anthropic-claude-fable-5-1-mythos-5-1-launch-2026 | Handy AI, Model Drop: Fable 5.1: https://handyai.substack.com/p/model-drop-fable-51