Claude 3.7 Sonnet vs Claude Fable 5.5 (Leaked Checkpoint)
A direct, empirical evaluation of Claude 3.7 Sonnet (Anthropic) against Claude Fable 5.5 (Leaked Checkpoint) (Anthropic).
Claude 3.7 Sonnet
Developed by Anthropic · Proprietary parameters
Enterprise software engineering, full-repo refactoring, and agentic workflows.
Claude Fable 5.5 (Leaked Checkpoint)
Developed by Anthropic · Proprietary Frontier Dynamic Reasoner parameters
Terminal-Bench 4.0 autonomous shell execution (78.5% target), recursive self-critique, and 1.5M-token repository refactoring.
| Metric | Claude 3.7 Sonnet | Claude Fable 5.5 (Leaked Checkpoint) |
|---|---|---|
| SWE-bench Verified | 70.3% | 84.2% |
| MMLU-Pro | 82.1% | 94.6% |
| MATH-500 | 96.2% | 98.4% |
| Context Window | 200,000 tokens | 1,500,000 tokens |
| Input Pricing (per 1M) | $3.00 | $3.50 |
| Output Pricing (per 1M) | $15.00 | $17.50 |