ModelWatch
← Catalog

Claude Fable 5.2

Anthropic's most capable widely released model. Best for the hardest reasoning and long-running agent work, where it can work for many minutes on one task.

reasoningagentscodinglong-context
Input price
$8 / 1M tokens
Output price
$40 / 1M tokens
Context window
1M tokens
Max output
128K tokens
Output speed
45 tokens/sseed
Time to first token
3.5sseed
Inputs
text, image, pdf
Open weights
No
API id
claude-fable-5-2
Released
Oct 15, 2026seed

Benchmarks

GPQA Diamond
89%
SWE-bench Verified
84%
MMLU-Pro
88%
AIME
96%
Arena Elo
1520

Benchmark scores are placeholders until sourced.

Compare with others