Anthropic released Claude Fable 5.1 (claude-fable-5-1) in September 2026; its lifecycle entry dates it September 1. The list price did not move: $10 per million input tokens and $50 per million output tokens. The number that did move is the one long-running agents spend most of their money on: cache reads now cost $0.25 per million tokens, 75% less than before.
What shipped
- Claude Fable 5.1,
claude-fable-5-1, active, with retirement "not sooner than September 1, 2027". - Claude Mythos 5.1,
claude-mythos-5-1, available only through Anthropic's trusted access programs (the Cyber Verification Program and the Life Sciences Verification Program), and for now only to a set of US organizations. - Pricing: $10 input, $50 output, $0.25 cache reads.
- A change for new API accounts: it is no longer possible for new accounts to manually edit prior context while keeping the transcript of Claude's earlier thinking. Existing accounts aren't affected yet, but Anthropic says the change will apply to future model releases.
Why the cache price is the headline
Take an agent session that re-sends a 300K-token prefix on each of 40 turns, with 2K new input tokens and 1K output tokens per turn. Almost all of the input is cached:
| Per session | Cached input | New input | Output |
|---|---|---|---|
| Tokens | 12M | 80K | 40K |
| At Fable 5.1 prices | $3.00 | $0.80 | $2.00 |
At the old cache-read price (four times higher) the cached part alone would have cost about $12. The cut turns cached context from the biggest line item into a minor one. That changes design decisions:
- Bigger stable prefixes are now affordable. Putting the full repository map, policy document or knowledge base in a cached prefix is cheaper than retrieving fragments on every turn.
- Cache hit rate becomes the metric to watch. A route with a poor hit rate pays $10 per million on the same tokens, 40 times the cached rate.
The integration change
The restriction on editing prior context while keeping Claude's thinking mostly affects custom integrations that rewrite history: summarizing old turns in place, or redacting a message after the fact. If your gateway or agent framework does that, plan for it now. Anthropic says it will apply to future models for all accounts.
Routing guidance
- Move long-session Fable lanes first. They gain the most from the cache price.
- Keep Fable 5 as the rollback. It stays active until at least June 9, 2027.
- Mythos 5.1 is not a general-purpose route. Access depends on verification, so don't build a default lane on it.
NeuronGate angle
NeuronGate records input and output tokens per request, so you can see which routes carry long prefixes and would benefit most from a cache-heavy model. Compare current prices in the model catalog, and see the docs for request details.
FAQ
Did Fable 5.1 get cheaper per token?
Input and output prices are unchanged at $10 and $50 per million. Cache reads dropped 75% to $0.25 per million.
Who can use Mythos 5.1?
Organizations admitted to Anthropic's Cyber Verification or Life Sciences Verification programs, currently a set of US organizations.
