Back to the index
Anthropic · Claude
No. 01

Claude Fable 5

Anthropic's most capable model — the reliability frontier for the hardest agentic work.

FrontierClosed weightsText, Image, Pdf
Field notes
Context window
1Mtokens
Max output
128Ktokens
Input price
$10/ 1M tokens
Output price
$50/ 1M tokens
Cached input
$1/ 1M tokens
Modalities
Text, Image, Pdf
Knowledge cutoff
Feb 2026
Released
Jun 2026
Approx. speed
Overview

What this model is

Claude Fable 5 is Anthropic's most capable widely-released model and the current reliability frontier for the hardest long-horizon agentic tasks. It sustains multi-hour autonomous coding sessions without derailing, corrupting a codebase, or looping, and holds coherence across its full million-token context. It is the model practitioners reach for when a complex agentic job absolutely must complete correctly.

Strengths

  • Best-in-class long-horizon agentic reliability
  • Elite real-world coding and multi-file refactoring
  • Sustained coherence over a 1M-token context
  • Exceptional instruction following and tool use
  • Strong scientific and graduate-level reasoning

Trade-offs

  • Highest price in the Claude lineup
  • Slower than smaller Claude models
  • No audio or video input
  • Output capped at 128K tokens
  • Overkill for routine or high-volume workloads

Best for

  • Autonomous multi-hour coding agents
  • Large-scale codebase migrations and refactors
  • Complex multi-step research and analysis
  • High-stakes agentic workflows
  • Hard reasoning and planning problems

Not ideal for

  • High-volume cheap classification
  • Latency-critical realtime chat
  • Simple one-shot tasks where cost matters
Capabilities

Capability profile

Normalized 0–100 scores, comparable across the whole catalog.

Reasoning97Coding98Math94Writing95Knowledge94Speed45Agentic98Vision89Multilingual92Long Context96
Reasoning
97
Coding
98
Math
94
Writing
95
Knowledge
94
Speed
45
Agentic
98
Vision
89
Multilingual
92
Long Context
96
Benchmarks

How it scores

Public benchmark results, with independent third-party results where available. Bars normalize percentages to 100 and Elo ratings to a 1500 ceiling.

Independent index
Artificial Analysis Intelligence Index
Composite of ~9–10 independent evals · Artificial Analysis
60/ 100
BenchmarkResult
Artificial Analysis Intelligence Indexindependent
GeneralArtificial Analysis
60%
SWE-bench Verifiedindependent
CodingSWE-bench
95%
SWE-bench Proindependent
CodingSWE-bench Pro
80%
τ-benchindependent
Agenticτ-bench
89.2%
LiveBenchindependent
GeneralLiveBench
78.3%
GPQA Diamondindependent
ReasoningIndependent aggregators
94.1%
Terminal-Bench Hardindependent
AgenticArtificial Analysis
62.9%
LMArena Eloindependent
GeneralLMArena
1525Elo
Alternatives

Comparable models

Models in a roughly similar class — worth weighing against this one.