Vendor-reported agentic science
WinnerTerminal-Bench-Science 0.1
Anthropic launch table. Standard error is reported at plus or minus 3.5 to 4.5 points per model.
SPIRITT
Claude Fable 5.1Claude Fable 5.1 stays sharp when the task gets long
Anthropic's September release leads the launch-day Artificial Analysis Intelligence Index at 66, reaches 91.4% on Terminal-Bench 2.1, and cuts cache-read pricing by 75%. In SPIRITT, give Fable 5.1 the browser, terminal, files, tools, and persistent context for the hardest multi-step work.
Claude Fable 5.1 across X
The official launch and independent evidence, followed by six concrete builds: kart racing, a living voxel kingdom, Minecraft, a fantasy Mount Fuji world, Rocket League, and a subway FPS.
Frontier reasoning for work that keeps unfolding
Fable 5.1 is strongest when the work spans tools, files, decisions, and verification loops. The evidence also shows a real cost and safeguard trade-off.
What Anthropic launched, and what it means in SPIRITT
Claude Fable 5.1 is Anthropic's generally available frontier model for demanding reasoning, long-horizon agentic coding, multistep research, and document, spreadsheet, and presentation work. Official documentation lists a 1M-token context window, 128K maximum output, text and image input, and adaptive thinking that is always on.
Artificial Analysis measured a launch-day Intelligence Index score of 66, the highest in its v4.1.1 snapshot, plus 91.4% on Terminal-Bench 2.1 and 1,853 Elo on GDPval-AA v2. Anthropic's own launch table reports leading results across terminal science, professional work, computer use, automation, and coding.
The $10 input and $50 output price per million tokens did not change from Fable 5. Cache reads fell to $0.25, but Artificial Analysis still measured max effort at roughly 20% more per task because Fable 5.1 used more output tokens. Its tested fallback also served about 4% of output tokens when safeguards intervened.
Claude Fable 5.1 is available through the Fable option in SPIRITT Workspaces. Put it on work where extra judgment, persistence, and verification are worth the premium, then keep consequential approvals and final review with a person.
66 AA Index91.4% Terminal-Bench 2.11M context128K output$0.25 cache readsAvailable in SPIRITT
AA Intelligence Index v4.1.1
WinnerArtificial Analysis launch snapshot, September 1, 2026. Fable 5.1 max with default fallback led at 66; fallback served about 4% of output tokens across the Index.
Where Fable 5.1 leads, and what that leadership costs
The ranking above is independent. The capability cards below use Anthropic's launch table except for the cost card, which uses Artificial Analysis measurements.
Vendor-reported terminal agents
WinnerTerminal-Bench 4.0
Anthropic launch table at each model's displayed high-capability setting.
Vendor-reported knowledge work
WinnerGDPval-AA v2 Elo
Anthropic launch table. The independent Artificial Analysis launch result also reports 1,853 Elo.
Vendor-reported computer use
WinnerOSWorld 2.0 partial score
Anthropic used the August 2026 task release. Earlier OSWorld 2.0 numbers are not directly comparable.
Vendor-reported reasoning
WinnerHumanity's Last Exam with tools
Anthropic launch table. The displayed frontier models are separated by less than 1.5 points.
Vendor-reported workflows
WinnerAutomationBench
Anthropic launch table. This is a strong relative gain, not broad task reliability.
Vendor-reported coding
WinnerCursorBench 3.2
Anthropic launch table. Cursor separately described Fable 5.1 as its strongest tested model on this benchmark.
Independent cost caveat
AA cost per Index task
Artificial Analysis launch snapshot. Lower is better; the cache cut did not offset higher output-token use at max effort.
Independent source: Artificial Analysis's September 1 Fable 5.1 launch evaluation. Vendor source: Anthropic's Fable 5.1 launch table and system card. Anthropic evaluated Fable 5.1 with production safeguards enabled, and both sources disclose fallback or safeguard effects. Scores can change with model effort, harness, fallback configuration, and benchmark revisions. Test your own highest-value workflows before switching production work.
How It Works
Choose Claude Fable 5.1 in the picker, then give it a finish line worthy of a frontier model
01
Open a workspace
Open a workspace and land in a fully equipped cloud computer: browser, files, terminal, integrations, and memory. No local setup. No thin chat box pretending to be an agent.
02
Choose Claude Fable 5.1
Open the SPIRITT model picker and select Claude Fable 5.1. Use it when long-horizon judgment, difficult debugging, deep research, or polished professional output justifies the premium tier.
03
Build or automate
Tell it what to ship or what to run. Fable 5.1 can code, call tools, drive the browser, coordinate multi-step work, and keep going while you step away. The point is not another chat window. It is an agentic environment where Fable 5.1 actually does the job.
Put Fable 5.1 on the work that keeps unfolding
Open a SPIRITT Workspace, choose Claude Fable 5.1, and give it the evidence, tools, constraints, and acceptance checks for a real end-to-end outcome.
Questions
Claude Fable 5.1 FAQ
01What is Claude Fable 5.1?+
Anthropic's September 2026 frontier model for demanding reasoning, long-horizon agentic coding, multistep research, and professional documents. It is the generally available configuration of the same underlying model used for Claude Mythos 5.1, which has different safeguards and restricted access.
02What is Fable 5.1 best at?+
Work that unfolds across many steps and rewards judgment: difficult repository changes, terminal tasks, research synthesis, computer use, document creation, and agent workflows that must verify their own progress instead of stopping after one answer.
03How strong is the independent evidence?+
Artificial Analysis put Fable 5.1 max at 66 on its launch-day Intelligence Index v4.1.1, the highest score in that snapshot. It also reported 91.4% on Terminal-Bench 2.1 and 1,853 Elo on GDPval-AA v2. The tested default fallback served about 4% of output tokens, so the result is for the routed production system rather than an isolated base model.
04How much does Claude Fable 5.1 cost?+
Anthropic lists $10 per million input tokens, $50 per million output tokens, and $0.25 per million cache-read tokens. Cache reads are 75% cheaper than Fable 5, but high output-token use can still make max-effort tasks more expensive overall.
05What fits in the context window?+
Official documentation lists a 1M-token context window and up to 128K output tokens. It accepts text and images and returns text. Long context is capacity, not a guarantee that every important detail will be recalled perfectly.
06What are the main limitations?+
It is a premium, slower model with adaptive thinking always on. Independent testing found higher cost per task than Fable 5 at max effort despite cheaper cache reads. Safeguards can refuse or route sensitive cyber and life-sciences work, forced tool use is not supported, and thinking blocks have migration constraints across models and edited conversation history.
07Is Claude Fable 5.1 available in SPIRITT Workspaces?+
Yes. Open a SPIRITT Workspace, choose the current Claude Fable option in the model picker, and use it with browser, terminal, files, tools, integrations, memory, and durable sessions.
Buy from builders who use what they sellBuilt using