Claude Fable 5.1, Anthropic's newest and most capable model, is now selectable across your CRHQ fleet. It is an upgrade to Fable 5, aimed squarely at the hardest, longest-running work, and it is meaningfully cheaper to run at scale.
What it is for
Fable 5.1 is built for tasks that run through many steps rather than answering once and stopping: deep investigations, large multi-stage builds, and autonomous agentic coding. Anthropic's framing is that it is better at working a problem all the way to the root cause instead of taking shortcuts that produce weaker results.
What is better
The numbers below are Anthropic's own reported results, not our measurements.
| Benchmark | What it measures | Fable 5 | Fable 5.1 |
|---|---|---|---|
| Terminal-Bench-Science | Long-horizon scientific problem solving | 24.7% | 52.6% |
| Terminal-Bench 4.0 | Agentic programming tasks | — | 55.8% |
The headline is Terminal-Bench-Science, where Fable 5.1 more than doubles its predecessor. It runs with a 1 million token context, thinks by default, and now lets you adjust effort mid-conversation, so you can dial how hard it works on a given step up or down without starting a new session.
Anthropic also reports fewer false positives from its safety classifiers, so legitimate work is less likely to be incorrectly declined.
Cheaper to run at scale
Fable 5.1 keeps Fable 5's per-token price of $10 in and $50 out per million tokens. What changed is the cost of cache reads, which is where long agentic sessions spend most of their tokens.
| Fable 5 | Fable 5.1 | |
|---|---|---|
| Input, per 1M tokens | $10.00 | $10.00 |
| Output, per 1M tokens | $50.00 | $50.00 |
| Cache read, per 1M tokens | $1.00 | $0.25 |
That is a 75% cut to the cache-read rate. Because a long-running agent re-reads its conversation prefix on every turn, this is the token type it uses most, and Anthropic reports agentic tasks running up to 45% cheaper overall as a result. Fable is still the premium tier, so use it deliberately for high-value work, but the economics of pointing it at a long autonomous job are now much better.
Pick your exact Fable version
CRHQ offers Fable the same way it offers Opus: as pinned versions plus a pass-through alias.
- Fable 5.1 and Fable 5 are pinned choices. Your selection is locked to that exact version and cannot change under you.
- Fable (latest) is the alias. It always follows whatever the provider currently ships as the default Fable, which today is Fable 5.1. If Anthropic moves that default later, the alias moves with it.
Pick a pinned version when you want certainty, or Fable (latest) when you want to ride the newest automatically.
How to use it
Open a session or agent model selector and choose Fable 5.1, or Fable (latest) to always track the newest. Set it as an agent's default where the work justifies the premium. Everything else, delegation, artifacts, skills, and tools, works exactly as before.
Availability
| Fact | Value |
|---|---|
| Availability | All CRHQ satellites |
| Live since | 2026-09-01 |
| Pricing | $10 in / $50 out per 1M tokens; cache reads $0.25 |
| Client action required | None. It is already live on your instance |