HomeAI models
Anthropic: Claude Fable 5.1
Anthropic
•
Claude Fable 5.1 (Max, Default Fallback)
•
Released 
September 2026

Anthropic
Claude Fable 5.1 (Max, Default Fallback)

Anthropic's frontier model for multi-hour autonomous agents, whole-codebase coding, and long-horizon knowledge work.

Modality:
Text
Image
Video
PDF
model ID
anthropic/claude-fable-5.1
Start for free
arrow black icon
View API Docs
arrow white icon

Output Speed *

71.89
tok/s

Intelligence Index *

53.4
/ 100

Context Window *

1000000
tokens

Input price

60
Anytoken

Output price

300
Anytoken
Claude Fable 5.1: Anthropic's Frontier Model for Long-Horizon Coding Agents Claude Fable 5.1 is Anthropic's most capable generally available model, sitting at the top of the Claude lineup above Opus, Sonnet, and Haiku. It's a Mythos-class model built for ambitious, long-running work: features that span an entire codebase, code review, and multi-hour autonomous agent sessions across many tools. Fable 5.1 improves most on agentic coding, long-running workflows, and knowledge work, and adds vision to interpret charts, diagrams, and dense PDFs. It suits teams running unattended agents rather than everyday chat or drafting, where lighter Claude models are more economical. Start building with Claude Fable 5.1 via the AnyAPI.ai API.

Performance

Where Fable 5.1 Pulls Ahead: Long, Multi-Step Agentic Work

Fable 5.1's largest gains are in long-horizon agentic tasks rather than short prompts. Anthropic reports 55.8% on Terminal-Bench 4.0 (up from Fable 5's 42.0%) and 52.6% on Terminal-Bench-Science 0.1, more than double its predecessor. The pattern matters because the model self-verifies, recovers from failed steps, and sustains multi-hour unattended runs. Independent testing by Snorkel AI found Fable 5.1 solved shared coding tasks using 58% fewer output tokens and 36% faster than Opus 5. In production, this means fewer failed attempts, less manual correction, and longer autonomous sessions before human intervention.

Benchmarks

Independent Benchmarks: Efficiency and Agentic Coding

On agentic coding, Fable 5.1 reached 73.4% on CursorBench 3.2 at max effort, described by Cursor as the most capable model it has run on that benchmark. Independent evaluator Snorkel AI compared Fable 5.1 against Opus 5 on frontier coding tasks: on problems both solved, Fable 5.1 used 58% fewer output tokens and finished 36% faster. Note that some widely circulated SWE-bench figures (95% Verified, 80% Pro) are Fable 5's numbers from June 2026, not confirmed for 5.1 — treat any 5.1 SWE-bench claim as unverified until independently published.

Output Speed

*
71.89
tok/s

Intelligence Index

*
53.4
/ 100

MMLU *

Broad world knowledge and problem-solving
0
%

GPQA *

PhD-level scientific reasoning across physics, biology, chemistry.
94
%

HLE *

Adherence to multi-step structured instructions.
59
%

LiveCodeBench *

Tool-calling reliability in long agentic loops.
0
%

Technical Specifications

What the model supports

Fable 5.1 accepts text, image, and file inputs and returns text. Its 1M-token context window is both the default and maximum, billed at standard per-token rates across the whole window — an advantage for whole-codebase and document-heavy work. Maximum output is 128,000 tokens per response. Adaptive thinking is always on and cannot be disabled; depth is set via an effort parameter (low, medium, high, xhigh, max). Because reasoning runs on every request, teams cannot opt out of thinking overhead, so effort tuning is central to controlling latency and cost.
Verified Specifications — 
Claude Fable 5.1 (Max, Default Fallback)
*
Input modalities
Text
Image
Video
PDF
output modalities
Text
Image
Context window
1000000
 tokens
Maximum output tokens
128000
Reasoning
Yes
Knowledge cutoff
September 2026
Pricing (standard)
60
 AnyTokens in
 / 
300
 AnyTokens out

Limitations & Trade-offs

Where Claude Fable 5.1 (Max, Default Fallback) falls short

1
Higher refusal rates. Fable 5.1 includes blocking classifiers for dual-use cybersecurity and life-sciences content, and Anthropic states refusal rates are materially higher than previous Claude models. Blocked requests return HTTP 200 with stop_reason: "refusal", and mid-stream refusals are billed for tokens generated before the block. Many bio/cyber-adjacent queries are rerouted to Opus models. Teams building security tooling, penetration testing, or life-sciences agents will hit frequent interruptions and must treat refusal as a primary response path, making Fable 5.1 unsuitable for some sensitive-domain work without Mythos access.
2
Premium uncached token pricing. Fable 5.1 sits at the top of the Claude family on list price, with input and output roughly twice Opus 5's rates. While cache reads are heavily discounted, uncached input and output remain expensive. For high-volume generation, everyday chat, or drafting, lighter Claude models such as Sonnet 5 are substantially more economical. Fable 5.1 must justify its premium through higher task completion, fewer output tokens, and longer autonomous runs rather than raw per-token price.
3
Forced reasoning overhead. Adaptive thinking is always on and cannot be disabled, unlike earlier Claude models where extended thinking was opt-in. Even the lowest effort level still incurs thinking. For latency-sensitive or simple high-throughput tasks, this overhead is unavoidable and adds cost and time, so applications that need near-instant short responses may prefer a non-reasoning or lighter model.
4
Breaking API changes and limits. Fable 5.1 introduces breaking changes versus Fable 5: forced tool_choice (any/tool) now returns a 400, and new API accounts can no longer manually edit prior context in multi-turn conversations (an anti-distillation measure). It is also not listed for the Batch API 300K extended-output beta available on Opus and Sonnet, capping output at 128K tokens. Teams migrating from Fable 5 must re-audit their messages arrays, tool logic, and evals before switching.

Best-Fit Workloads

Where this model earns its place

01

Long-horizon autonomous coding agents

‍
Fable 5.1 is Anthropic's most capable model for ambitious coding: features spanning an entire codebase, code review, performance work, and multi-day autonomous sessions. It writes its own tests, self-verifies, and recovers from failed steps. Anthropic reports 55.8% on Terminal-Bench 4.0 and 73.4% on CursorBench 3.2 at max effort. In agent harnesses like Cursor or Windsurf, this translates to fewer failed attempts and longer runs before human intervention. Cache-read discounts particularly benefit these repetitive-context agentic loops.

02

Agentic scientific and research automation

‍
Fable 5.1 more than doubled Fable 5 on Terminal-Bench-Science 0.1 (24.7% to 52.6%) and roughly doubled AutomationBench (17.1% to 31.4%), which measures multi-step agentic automation chaining tool calls across systems. This suits research pipelines and knowledge-work automation that run for hours across many applications. Note that safeguards may reroute or refuse bio-adjacent queries, so research in restricted domains may require handling refusals or Mythos access.

03
Document-heavy analysis with vision

‍
Fable 5.1 brings strong vision to dense filings, charts, diagrams, and tables nested in files and PDFs, improving document-heavy work in finance, legal, analytics, and architecture. Combined with the 1M-token context window billed at standard rates across the whole window, it can ingest large document sets in a single pass. Teams processing complex financial or legal documents benefit from its combined vision and long-context handling.

04

Whole-codebase refactoring and migration

‍
The 1M-token context lets Fable 5.1 hold large portions of a codebase in a single request, and OpenRouter notes its biggest gains include long code refactors and front-end/visual code generation. It uses vision to check outputs against original designs. For large migrations or refactors that previously required repeated human check-ins, its self-correction and long-context capacity reduce fragmentation, though premium uncached pricing means well-structured, cache-friendly prompting matters for cost.

Pricing in anytokens via AnyAPI
Input
60
₳
Output
300
₳
Cache write
1.5
₳
Cache read
75
₳
Get API access
arrow black icon
Read Docs

on this page

PerformanceBenchmarksTechnical SpecificationsLimitations & Trade-offsBest-Fit WorkloadsIntegrationFAQ

Integration

Access Claude Fable 5.1 (Max, Default Fallback) via AnyAPI.ai

Access Claude Fable 5.1 (Max, Default Fallback) through AnyAPI.ai using a unified API built for multi-model AI applications. Integrate Claude Fable 5.1 (Max, Default Fallback) without maintaining a separate provider-specific connection, and keep the flexibility to test, switch, or combine models as your application requirements evolve.

01

One API integration

Access Claude Fable 5.1 (Max, Default Fallback) and other AI models through the same API workflow instead of maintaining separate integrations for every provider.

02

Easy model switching

Test Claude Fable 5.1 (Max, Default Fallback) against alternative models or switch models as your performance, capability, or cost requirements change without rebuilding your application around another provider API.

03

Flexible for production

Use Claude Fable 5.1 (Max, Default Fallback) from experimentation through production while keeping your AI stack flexible as workloads, traffic, and model requirements evolve.

04

Multi-model applications

Use Claude Fable 5.1 (Max, Default Fallback) for the workloads where it performs best and combine it with other models for tasks that require different capabilities, performance, or efficiency.

Start free
arrow black icon
Read Docs

Frequently Asked Questions

Answers to common questions about integrating and using this AI model via AnyAPI.ai

Claude Fable 5.1 has a 1,000,000-token context window, which is both the default and maximum, billed at standard per-token rates across the entire window. Maximum output is 128,000 tokens per response. Unlike Opus 5 and Sonnet 5, it is not listed for the Batch API 300K extended-output beta.

Yes, but reasoning cannot be turned off. Fable 5.1 uses adaptive thinking that is always on, unlike earlier Claude models where extended thinking was opt-in. You control depth via an effort parameter with five levels — low, medium, high, xhigh, and max — with a default of high, trading latency and cost for reasoning depth.

Anthropic positions Fable 5.1 as leading Opus 5 on every benchmark it published, and independent testing found Fable 5.1 solved shared coding tasks with 58% fewer output tokens and 36% faster than Opus 5. However, Fable 5.1 sits higher on list price. Opus 5 is the more economical default for everyday reasoning and chat.

Fable 5.1 includes blocking classifiers for dual-use cybersecurity and life-sciences content, and Anthropic notes refusal rates are materially higher than earlier Claude models. Blocked requests return HTTP 200 with stop_reason: "refusal". Many bio- and cyber-adjacent queries are rerouted to Opus models. Developers should handle refusal as a primary response path.

Claude Fable 5.1 accepts text, image, and file inputs, including PDFs, and returns text output. Its vision is designed for dense filings, charts, diagrams, and tables nested in documents, and it can use vision to check its own coding outputs against a design or goal. It does not generate images or audio.

* Benchmark data source: Artificial Analysis artificialanalysis.ai