Anthropic cut prices 25 percent with Claude Fable 5.1 and Mythos 5.1

Anthropic released Claude Fable 5.1 and Claude Mythos 5.1 on 1 September 2026. Fable 5.1 is the generally available model for coding and knowledge work. Mythos 5.1 is the same model with relaxed safeguards for vetted cybersecurity and life sciences work in the United States, reached through Anthropic's Cyber Verification Program and Life Sciences Verification Program.

What the numbers look like

Fable 5.1 is priced at $10 per million input tokens and $50 per million output tokens. Cache reads are priced at $0.25 per million tokens, which Anthropic says is 75 percent lower than before. On a typical workload the company estimates around 25 percent lower cost against Fable 5, rising to around 45 percent on agentic tasks that use the cache heavily.

The benchmarks on Anthropic's own page show a wide range. Fable 5.1 scores 55.8 percent on Terminal-Bench 4.0 against Fable 5's 42.0 percent, 52.6 percent on Terminal-Bench-Science 0.1 against 24.7 percent, and 31.4 percent on AutomationBench against 17.1 percent. On OSWorld 2.0 the partial score goes from 72.9 percent to 77.9 percent, and the strict score from 36.1 percent to 41.7 percent. On CursorBench 3.2.0 it reaches 73.4 percent, up from 70.5 percent. Mythos 5.1 reaches 60.9 percent on Terminal-Bench 4.0, up from Mythos 5's 37.3 percent.

How the tools change

Anthropic says Fable 5.1 is available immediately on Amazon Web Services, Google Cloud and Microsoft Azure, and developers access it on the Claude API through the identifier claude-fable-5-1. The page also notes that Fable 5.1 defaults to High effort in Claude Code and Medium in Claude Cowork and on Claude.ai. Mythos 5.1 is restricted to United States organisations that pass one of the two verification programs.

The GitHub release for the Claude Code command line client the same day lists Claude Fable 5.1 as the new default Fable model and adds a Containment Escape rule to auto mode, so cloud metadata credential fetches, egress evasion and cross-tenant reach are no longer auto approved unless the environment marks them expected.

The price drop is the change that appears on the invoice. A team already routing production work through Claude on the API can rerun this month's spend against the new rates and see the difference without any code change: the cache read cut alone is worth a lot on retrieval augmented workloads where the same context is read on every turn. The 25 percent typical figure and 45 percent agentic figure are Anthropic's own, so worth measuring against a real workload rather than trusting.

The Containment Escape rule in the Claude Code release is the small operational item that is easy to miss. It applies in auto mode only, and it targets three specific behaviours: reaching cloud instance metadata, evading egress rules, and touching another tenant. If your build pipeline runs Claude Code with auto mode against a new cloud environment, the auto approval you rely on will now fail unless the environment sets the expected marker. It is better to find this in a sandbox than during a production run.

Source

Primary source: Claude Fable 5.1 and Claude Mythos 5.1, Anthropic, 1 September 2026.